update ui tacotron

TTS/.models.json

@@ -0,0 +1,515 @@
+{
+    "tts_models": {
+        "multilingual":{
+            "multi-dataset":{
+                "your_tts":{
+                    "description": "Your TTS model accompanying the paper https://arxiv.org/abs/2112.02418",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--multilingual--multi-dataset--your_tts.zip",
+                    "default_vocoder": null,
+                    "commit": "e9a1953e",
+                    "license": "CC BY-NC-ND 4.0",
+                    "contact": "[email protected]"
+                }
+            }
+        },
+        "en": {
+            "ek1": {
+                "tacotron2": {
+                    "description": "EK1 en-rp tacotron2 by NMStoker",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ek1--tacotron2.zip",
+                    "default_vocoder": "vocoder_models/en/ek1/wavegrad",
+                    "commit": "c802255",
+                    "license": "apache 2.0"
+                }
+            },
+            "ljspeech": {
+                "tacotron2-DDC": {
+                    "description": "Tacotron2 with Double Decoder Consistency.",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--tacotron2-DDC.zip",
+                    "default_vocoder": "vocoder_models/en/ljspeech/hifigan_v2",
+                    "commit": "bae2ad0f",
+                    "author": "Eren Gölge @erogol",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                },
+                "tacotron2-DDC_ph": {
+                    "description": "Tacotron2 with Double Decoder Consistency with phonemes.",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--tacotron2-DDC_ph.zip",
+                    "default_vocoder": "vocoder_models/en/ljspeech/univnet",
+                    "commit": "3900448",
+                    "author": "Eren Gölge @erogol",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                },
+                "glow-tts": {
+                    "description": "",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--glow-tts.zip",
+                    "stats_file": null,
+                    "default_vocoder": "vocoder_models/en/ljspeech/multiband-melgan",
+                    "commit": "",
+                    "author": "Eren Gölge @erogol",
+                    "license": "MPL",
+                    "contact": "[email protected]"
+                },
+                "speedy-speech": {
+                    "description": "Speedy Speech model trained on LJSpeech dataset using the Alignment Network for learning the durations.",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--speedy-speech.zip",
+                    "stats_file": null,
+                    "default_vocoder": "vocoder_models/en/ljspeech/hifigan_v2",
+                    "commit": "4581e3d",
+                    "author": "Eren Gölge @erogol",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                },
+                "tacotron2-DCA": {
+                    "description": "",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--tacotron2-DCA.zip",
+                    "default_vocoder": "vocoder_models/en/ljspeech/multiband-melgan",
+                    "commit": "",
+                    "author": "Eren Gölge @erogol",
+                    "license": "MPL",
+                    "contact": "[email protected]"
+                },
+                "vits": {
+                    "description": "VITS is an End2End TTS model trained on LJSpeech dataset with phonemes.",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--vits.zip",
+                    "default_vocoder": null,
+                    "commit": "3900448",
+                    "author": "Eren Gölge @erogol",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                },
+                "fast_pitch": {
+                    "description": "FastPitch model trained on LJSpeech using the Aligner Network",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--fast_pitch.zip",
+                    "default_vocoder": "vocoder_models/en/ljspeech/hifigan_v2",
+                    "commit": "b27b3ba",
+                    "author": "Eren Gölge @erogol",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                }
+            },
+            "vctk": {
+                "vits": {
+                    "description": "VITS End2End TTS model trained on VCTK dataset with 109 different speakers with EN accent.",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--vctk--vits.zip",
+                    "default_vocoder": null,
+                    "commit": "3900448",
+                    "author": "Eren @erogol",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                },
+                "fast_pitch":{
+                    "description": "FastPitch model trained on VCTK dataseset.",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--vctk--fast_pitch.zip",
+                    "default_vocoder": null,
+                    "commit": "bdab788d",
+                    "author": "Eren @erogol",
+                    "license": "CC BY-NC-ND 4.0",
+                    "contact": "[email protected]"
+                }
+            },
+            "sam": {
+                "tacotron-DDC": {
+                    "description": "Tacotron2 with Double Decoder Consistency trained with Aceenture's Sam dataset.",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--sam--tacotron-DDC.zip",
+                    "default_vocoder": "vocoder_models/en/sam/hifigan_v2",
+                    "commit": "bae2ad0f",
+                    "author": "Eren Gölge @erogol",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                }
+            },
+            "blizzard2013": {
+                "capacitron-t2-c50": {
+                    "description": "Capacitron additions to Tacotron 2 with Capacity at 50 as in https://arxiv.org/pdf/1906.03402.pdf",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.7.0_models/tts_models--en--blizzard2013--capacitron-t2-c50.zip",
+                    "commit": "d6284e7",
+                    "default_vocoder": "vocoder_models/en/blizzard2013/hifigan_v2",
+                    "author": "Adam Froghyar @a-froghyar",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                },
+                "capacitron-t2-c150_v2": {
+                    "description": "Capacitron additions to Tacotron 2 with Capacity at 150 as in https://arxiv.org/pdf/1906.03402.pdf",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.7.1_models/tts_models--en--blizzard2013--capacitron-t2-c150_v2.zip",
+                    "commit": "a67039d",
+                    "default_vocoder": "vocoder_models/en/blizzard2013/hifigan_v2",
+                    "author": "Adam Froghyar @a-froghyar",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                }
+            }
+        },
+        "es": {
+            "mai": {
+                "tacotron2-DDC": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--es--mai--tacotron2-DDC.zip",
+                    "default_vocoder": "vocoder_models/universal/libri-tts/fullband-melgan",
+                    "commit": "",
+                    "author": "Eren Gölge @erogol",
+                    "license": "MPL",
+                    "contact": "[email protected]"
+                }
+            }
+        },
+        "fr": {
+            "mai": {
+                "tacotron2-DDC": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--fr--mai--tacotron2-DDC.zip",
+                    "default_vocoder": "vocoder_models/universal/libri-tts/fullband-melgan",
+                    "commit": "",
+                    "author": "Eren Gölge @erogol",
+                    "license": "MPL",
+                    "contact": "[email protected]"
+                }
+            }
+        },
+        "uk":{
+            "mai": {
+                "glow-tts": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--uk--mai--glow-tts.zip",
+                    "author":"@robinhad",
+                    "commit": "bdab788d",
+                    "license": "MIT",
+                    "contact": "",
+                    "default_vocoder": "vocoder_models/uk/mai/multiband-melgan"
+                }
+            }
+        },
+        "zh-CN": {
+            "baker": {
+                "tacotron2-DDC-GST": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--zh-CN--baker--tacotron2-DDC-GST.zip",
+                    "commit": "unknown",
+                    "author": "@kirianguiller",
+                    "license": "apache 2.0",
+                    "default_vocoder": null
+                }
+            }
+        },
+        "nl": {
+            "mai": {
+                "tacotron2-DDC": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--nl--mai--tacotron2-DDC.zip",
+                    "author": "@r-dh",
+                    "license": "apache 2.0",
+                    "default_vocoder": "vocoder_models/nl/mai/parallel-wavegan",
+                    "stats_file": null,
+                    "commit": "540d811"
+                }
+            }
+        },
+        "de": {
+            "thorsten": {
+                "tacotron2-DCA": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--de--thorsten--tacotron2-DCA.zip",
+                    "default_vocoder": "vocoder_models/de/thorsten/fullband-melgan",
+                    "author": "@thorstenMueller",
+                    "license": "apache 2.0",
+                    "commit": "unknown"
+                },
+                "vits": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.7.0_models/tts_models--de--thorsten--vits.zip",
+                    "default_vocoder": null,
+                    "author": "@thorstenMueller",
+                    "license": "apache 2.0",
+                    "commit": "unknown"
+                },
+                "tacotron2-DDC": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--de--thorsten--tacotron2-DDC.zip",
+                    "default_vocoder": "vocoder_models/de/thorsten/hifigan_v1",
+                    "description": "Thorsten-Dec2021-22k-DDC",
+                    "author": "@thorstenMueller",
+                    "license": "apache 2.0",
+                    "commit": "unknown"
+                }
+            }
+        },
+        "ja": {
+            "kokoro": {
+                "tacotron2-DDC": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--ja--kokoro--tacotron2-DDC.zip",
+                    "default_vocoder": "vocoder_models/ja/kokoro/hifigan_v1",
+                    "description": "Tacotron2 with Double Decoder Consistency trained with Kokoro Speech Dataset.",
+                    "author": "@kaiidams",
+                    "license": "apache 2.0",
+                    "commit": "401fbd89"
+                }
+            }
+        },
+        "tr":{
+            "common-voice": {
+                "glow-tts":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--tr--common-voice--glow-tts.zip",
+                    "default_vocoder": "vocoder_models/tr/common-voice/hifigan",
+                    "license": "MIT",
+                    "description": "Turkish GlowTTS model using an unknown speaker from the Common-Voice dataset.",
+                    "author": "Fatih Akademi",
+                    "commit": null
+                }
+            }
+        },
+        "it": {
+            "mai_female": {
+                "glow-tts":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--it--mai_female--glow-tts.zip",
+                    "default_vocoder": null,
+                    "description": "GlowTTS model as explained on https://github.com/coqui-ai/TTS/issues/1148.",
+                    "author": "@nicolalandro",
+                    "license": "apache 2.0",
+                    "commit": null
+                },
+                "vits":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--it--mai_female--vits.zip",
+                    "default_vocoder": null,
+                    "description": "GlowTTS model as explained on https://github.com/coqui-ai/TTS/issues/1148.",
+                    "author": "@nicolalandro",
+                    "license": "apache 2.0",
+                    "commit": null
+                }
+            },
+            "mai_male": {
+                "glow-tts":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--it--mai_male--glow-tts.zip",
+                    "default_vocoder": null,
+                    "description": "GlowTTS model as explained on https://github.com/coqui-ai/TTS/issues/1148.",
+                    "author": "@nicolalandro",
+                    "license": "apache 2.0",
+                    "commit": null
+                },
+                "vits":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--it--mai_male--vits.zip",
+                    "default_vocoder": null,
+                    "description": "GlowTTS model as explained on https://github.com/coqui-ai/TTS/issues/1148.",
+                    "author": "@nicolalandro",
+                    "license": "apache 2.0",
+                    "commit": null
+                }
+            }
+        },
+        "ewe": {
+            "openbible": {
+                "vits":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--ewe--openbible--vits.zip",
+                    "default_vocoder": null,
+                    "license": "CC-BY-SA 4.0",
+                    "description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
+                    "author": "@coqui_ai",
+                    "commit": "1b22f03"
+                }
+            }
+        },
+        "hau": {
+            "openbible": {
+                "vits":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--hau--openbible--vits.zip",
+                    "default_vocoder": null,
+                    "license": "CC-BY-SA 4.0",
+                    "description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
+                    "author": "@coqui_ai",
+                    "commit": "1b22f03"
+                }
+            }
+        },
+        "lin": {
+            "openbible": {
+                "vits":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--lin--openbible--vits.zip",
+                    "default_vocoder": null,
+                    "license": "CC-BY-SA 4.0",
+                    "description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
+                    "author": "@coqui_ai",
+                    "commit": "1b22f03"
+                }
+            }
+        },
+        "tw_akuapem": {
+            "openbible": {
+                "vits":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--tw_akuapem--openbible--vits.zip",
+                    "default_vocoder": null,
+                    "license": "CC-BY-SA 4.0",
+                    "description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
+                    "author": "@coqui_ai",
+                    "commit": "1b22f03"
+                }
+            }
+        },
+        "tw_asante": {
+            "openbible": {
+                "vits":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--tw_asante--openbible--vits.zip",
+                    "default_vocoder": null,
+                    "license": "CC-BY-SA 4.0",
+                    "description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
+                    "author": "@coqui_ai",
+                    "commit": "1b22f03"
+                }
+            }
+        },
+        "yor": {
+            "openbible": {
+                "vits":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--yor--openbible--vits.zip",
+                    "default_vocoder": null,
+                    "license": "CC-BY-SA 4.0",
+                    "description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
+                    "author": "@coqui_ai",
+                    "commit": "1b22f03"
+                }
+            }
+        }
+    },
+    "vocoder_models": {
+        "universal": {
+            "libri-tts": {
+                "wavegrad": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--universal--libri-tts--wavegrad.zip",
+                    "commit": "ea976b0",
+                    "author": "Eren Gölge @erogol",
+                    "license": "MPL",
+                    "contact": "[email protected]"
+                },
+                "fullband-melgan": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--universal--libri-tts--fullband-melgan.zip",
+                    "commit": "4132240",
+                    "author": "Eren Gölge @erogol",
+                    "license": "MPL",
+                    "contact": "[email protected]"
+                }
+            }
+        },
+        "en": {
+            "ek1": {
+                "wavegrad": {
+                    "description": "EK1 en-rp wavegrad by NMStoker",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--ek1--wavegrad.zip",
+                    "commit": "c802255",
+                    "license": "apache 2.0"
+                }
+            },
+            "ljspeech": {
+                "multiband-melgan": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--ljspeech--multiband-melgan.zip",
+                    "commit": "ea976b0",
+                    "author": "Eren Gölge @erogol",
+                    "license": "MPL",
+                    "contact": "[email protected]"
+                },
+                "hifigan_v2": {
+                    "description": "HiFiGAN_v2 LJSpeech vocoder from https://arxiv.org/abs/2010.05646.",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--ljspeech--hifigan_v2.zip",
+                    "commit": "bae2ad0f",
+                    "author": "@erogol",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                },
+                "univnet": {
+                    "description": "UnivNet model finetuned on TacotronDDC_ph spectrograms for better compatibility.",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--ljspeech--univnet_v2.zip",
+                    "commit": "4581e3d",
+                    "author": "Eren @erogol",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                }
+            },
+            "blizzard2013": {
+                "hifigan_v2": {
+                    "description": "HiFiGAN_v2 LJSpeech vocoder from https://arxiv.org/abs/2010.05646.",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.7.0_models/vocoder_models--en--blizzard2013--hifigan_v2.zip",
+                    "commit": "d6284e7",
+                    "author": "Adam Froghyar @a-froghyar",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                }
+            },
+            "vctk": {
+                "hifigan_v2": {
+                    "description": "Finetuned and intended to be used with tts_models/en/vctk/sc-glow-tts",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--vctk--hifigan_v2.zip",
+                    "commit": "2f07160",
+                    "author": "Edresson Casanova",
+                    "license": "apache 2.0",
+                    "contact": ""
+                }
+            },
+            "sam": {
+                "hifigan_v2": {
+                    "description": "Finetuned and intended to be used with tts_models/en/sam/tacotron_DDC",
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--sam--hifigan_v2.zip",
+                    "commit": "2f07160",
+                    "author": "Eren Gölge @erogol",
+                    "license": "apache 2.0",
+                    "contact": "[email protected]"
+                }
+            }
+        },
+        "nl": {
+            "mai": {
+                "parallel-wavegan": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--nl--mai--parallel-wavegan.zip",
+                    "author": "@r-dh",
+                    "license": "apache 2.0",
+                    "commit": "unknown"
+                }
+            }
+        },
+        "de": {
+            "thorsten": {
+                "wavegrad": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--de--thorsten--wavegrad.zip",
+                    "author": "@thorstenMueller",
+                    "license": "apache 2.0",
+                    "commit": "unknown"
+                },
+                "fullband-melgan": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--de--thorsten--fullband-melgan.zip",
+                    "author": "@thorstenMueller",
+                    "license": "apache 2.0",
+                    "commit": "unknown"
+                },
+                "hifigan_v1": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/vocoder_models--de--thorsten--hifigan_v1.zip",
+                    "description": "HifiGAN vocoder model for Thorsten Neutral Dec2021 22k Samplerate Tacotron2 DDC model",
+                    "author": "@thorstenMueller",
+                    "license": "apache 2.0",
+                    "commit": "unknown"
+                }
+            }
+        },
+        "ja": {
+            "kokoro": {
+                "hifigan_v1": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--ja--kokoro--hifigan_v1.zip",
+                    "description": "HifiGAN model trained for kokoro dataset by @kaiidams",
+                    "author": "@kaiidams",
+                    "license": "apache 2.0",
+                    "commit": "3900448"
+                }
+            }
+        },
+        "uk": {
+            "mai": {
+                "multiband-melgan": {
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--uk--mai--multiband-melgan.zip",
+                    "author":"@robinhad",
+                    "commit": "bdab788d",
+                    "license": "MIT",
+                    "contact": ""
+                }
+            }
+        },
+        "tr":{
+            "common-voice": {
+                "hifigan":{
+                    "github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--tr--common-voice--hifigan.zip",
+                    "description": "HifiGAN model using an unknown speaker from the Common-Voice dataset.",
+                    "author": "Fatih Akademi",
+                    "license": "MIT",
+                    "commit": null
+                }
+            }
+        }
+    }
+}

TTS/VERSION

@@ -0,0 +1 @@
+0.8.0

TTS/__init__.py

@@ -0,0 +1,6 @@
+import os
+
+with open(os.path.join(os.path.dirname(__file__), "VERSION"), "r", encoding="utf-8") as f:
+    version = f.read().strip()
+
+__version__ = version

TTS/bin/.ipynb_checkpoints/find_unique_chars-checkpoint.py

@@ -0,0 +1,45 @@
+"""Find all the unique characters in a dataset"""
+import argparse
+from argparse import RawTextHelpFormatter
+
+from TTS.config import load_config
+from TTS.tts.datasets import load_tts_samples
+
+
+def main():
+    # pylint: disable=bad-option-value
+    parser = argparse.ArgumentParser(
+        description="""Find all the unique characters or phonemes in a dataset.\n\n"""
+        """
+    Example runs:
+
+    python TTS/bin/find_unique_chars.py --config_path config.json
+    """,
+        formatter_class=RawTextHelpFormatter,
+    )
+    parser.add_argument("--config_path", type=str, help="Path to dataset config file.", required=True)
+    args = parser.parse_args()
+
+    c = load_config(args.config_path)
+
+    # load all datasets
+    train_items, eval_items = load_tts_samples(
+        c.datasets, eval_split=True, eval_split_max_size=c.eval_split_max_size, eval_split_size=c.eval_split_size
+    )
+
+    items = train_items + eval_items
+
+    texts = "".join(item["text"] for item in items)
+    chars = set(texts)
+    lower_chars = filter(lambda c: c.islower(), chars)
+    chars_force_lower = [c.lower() for c in chars]
+    chars_force_lower = set(chars_force_lower)
+
+    print(f" > Number of unique characters: {len(chars)}")
+    print(f" > Unique characters: {''.join(sorted(chars))}")
+    print(f" > Unique lower characters: {''.join(sorted(lower_chars))}")
+    print(f" > Unique all forced to lower characters: {''.join(sorted(chars_force_lower))}")
+
+
+if __name__ == "__main__":
+    main()

TTS/bin/collect_env_info.py

@@ -0,0 +1,48 @@
+"""Get detailed info about the working environment."""
+import os
+import platform
+import sys
+
+import numpy
+import torch
+
+sys.path += [os.path.abspath(".."), os.path.abspath(".")]
+import json
+
+import TTS
+
+
+def system_info():
+    return {
+        "OS": platform.system(),
+        "architecture": platform.architecture(),
+        "version": platform.version(),
+        "processor": platform.processor(),
+        "python": platform.python_version(),
+    }
+
+
+def cuda_info():
+    return {
+        "GPU": [torch.cuda.get_device_name(i) for i in range(torch.cuda.device_count())],
+        "available": torch.cuda.is_available(),
+        "version": torch.version.cuda,
+    }
+
+
+def package_info():
+    return {
+        "numpy": numpy.__version__,
+        "PyTorch_version": torch.__version__,
+        "PyTorch_debug": torch.version.debug,
+        "TTS": TTS.__version__,
+    }
+
+
+def main():
+    details = {"System": system_info(), "CUDA": cuda_info(), "Packages": package_info()}
+    print(json.dumps(details, indent=4, sort_keys=True))
+
+
+if __name__ == "__main__":
+    main()

TTS/bin/compute_attention_masks.py

@@ -0,0 +1,165 @@
+import argparse
+import importlib
+import os
+from argparse import RawTextHelpFormatter
+
+import numpy as np
+import torch
+from torch.utils.data import DataLoader
+from tqdm import tqdm
+
+from TTS.config import load_config
+from TTS.tts.datasets.TTSDataset import TTSDataset
+from TTS.tts.models import setup_model
+from TTS.tts.utils.text.characters import make_symbols, phonemes, symbols
+from TTS.utils.audio import AudioProcessor
+from TTS.utils.io import load_checkpoint
+
+if __name__ == "__main__":
+    # pylint: disable=bad-option-value
+    parser = argparse.ArgumentParser(
+        description="""Extract attention masks from trained Tacotron/Tacotron2 models.
+These masks can be used for different purposes including training a TTS model with a Duration Predictor.\n\n"""
+        """Each attention mask is written to the same path as the input wav file with ".npy" file extension.
+(e.g. path/bla.wav (wav file) --> path/bla.npy (attention mask))\n"""
+        """
+Example run:
+    CUDA_VISIBLE_DEVICE="0" python TTS/bin/compute_attention_masks.py
+        --model_path /data/rw/home/Models/ljspeech-dcattn-December-14-2020_11+10AM-9d0e8c7/checkpoint_200000.pth
+        --config_path /data/rw/home/Models/ljspeech-dcattn-December-14-2020_11+10AM-9d0e8c7/config.json
+        --dataset_metafile metadata.csv
+        --data_path /root/LJSpeech-1.1/
+        --batch_size 32
+        --dataset ljspeech
+        --use_cuda True
+""",
+        formatter_class=RawTextHelpFormatter,
+    )
+    parser.add_argument("--model_path", type=str, required=True, help="Path to Tacotron/Tacotron2 model file ")
+    parser.add_argument(
+        "--config_path",
+        type=str,
+        required=True,
+        help="Path to Tacotron/Tacotron2 config file.",
+    )
+    parser.add_argument(
+        "--dataset",
+        type=str,
+        default="",
+        required=True,
+        help="Target dataset processor name from TTS.tts.dataset.preprocess.",
+    )
+
+    parser.add_argument(
+        "--dataset_metafile",
+        type=str,
+        default="",
+        required=True,
+        help="Dataset metafile inclusing file paths with transcripts.",
+    )
+    parser.add_argument("--data_path", type=str, default="", help="Defines the data path. It overwrites config.json.")
+    parser.add_argument("--use_cuda", type=bool, default=False, help="enable/disable cuda.")
+
+    parser.add_argument(
+        "--batch_size", default=16, type=int, help="Batch size for the model. Use batch_size=1 if you have no CUDA."
+    )
+    args = parser.parse_args()
+
+    C = load_config(args.config_path)
+    ap = AudioProcessor(**C.audio)
+
+    # if the vocabulary was passed, replace the default
+    if "characters" in C.keys():
+        symbols, phonemes = make_symbols(**C.characters)
+
+    # load the model
+    num_chars = len(phonemes) if C.use_phonemes else len(symbols)
+    # TODO: handle multi-speaker
+    model = setup_model(C)
+    model, _ = load_checkpoint(model, args.model_path, args.use_cuda, True)
+
+    # data loader
+    preprocessor = importlib.import_module("TTS.tts.datasets.formatters")
+    preprocessor = getattr(preprocessor, args.dataset)
+    meta_data = preprocessor(args.data_path, args.dataset_metafile)
+    dataset = TTSDataset(
+        model.decoder.r,
+        C.text_cleaner,
+        compute_linear_spec=False,
+        ap=ap,
+        meta_data=meta_data,
+        characters=C.characters if "characters" in C.keys() else None,
+        add_blank=C["add_blank"] if "add_blank" in C.keys() else False,
+        use_phonemes=C.use_phonemes,
+        phoneme_cache_path=C.phoneme_cache_path,
+        phoneme_language=C.phoneme_language,
+        enable_eos_bos=C.enable_eos_bos_chars,
+    )
+
+    dataset.sort_and_filter_items(C.get("sort_by_audio_len", default=False))
+    loader = DataLoader(
+        dataset,
+        batch_size=args.batch_size,
+        num_workers=4,
+        collate_fn=dataset.collate_fn,
+        shuffle=False,
+        drop_last=False,
+    )
+
+    # compute attentions
+    file_paths = []
+    with torch.no_grad():
+        for data in tqdm(loader):
+            # setup input data
+            text_input = data[0]
+            text_lengths = data[1]
+            linear_input = data[3]
+            mel_input = data[4]
+            mel_lengths = data[5]
+            stop_targets = data[6]
+            item_idxs = data[7]
+
+            # dispatch data to GPU
+            if args.use_cuda:
+                text_input = text_input.cuda()
+                text_lengths = text_lengths.cuda()
+                mel_input = mel_input.cuda()
+                mel_lengths = mel_lengths.cuda()
+
+            model_outputs = model.forward(text_input, text_lengths, mel_input)
+
+            alignments = model_outputs["alignments"].detach()
+            for idx, alignment in enumerate(alignments):
+                item_idx = item_idxs[idx]
+                # interpolate if r > 1
+                alignment = (
+                    torch.nn.functional.interpolate(
+                        alignment.transpose(0, 1).unsqueeze(0),
+                        size=None,
+                        scale_factor=model.decoder.r,
+                        mode="nearest",
+                        align_corners=None,
+                        recompute_scale_factor=None,
+                    )
+                    .squeeze(0)
+                    .transpose(0, 1)
+                )
+                # remove paddings
+                alignment = alignment[: mel_lengths[idx], : text_lengths[idx]].cpu().numpy()
+                # set file paths
+                wav_file_name = os.path.basename(item_idx)
+                align_file_name = os.path.splitext(wav_file_name)[0] + "_attn.npy"
+                file_path = item_idx.replace(wav_file_name, align_file_name)
+                # save output
+                wav_file_abs_path = os.path.abspath(item_idx)
+                file_abs_path = os.path.abspath(file_path)
+                file_paths.append([wav_file_abs_path, file_abs_path])
+                np.save(file_path, alignment)
+
+        # ourput metafile
+        metafile = os.path.join(args.data_path, "metadata_attn_mask.txt")
+
+        with open(metafile, "w", encoding="utf-8") as f:
+            for p in file_paths:
+                f.write(f"{p[0]}|{p[1]}\n")
+        print(f" >> Metafile created: {metafile}")

TTS/bin/compute_embeddings.py

@@ -0,0 +1,131 @@
+import argparse
+import os
+from argparse import RawTextHelpFormatter
+
+import torch
+from tqdm import tqdm
+
+from TTS.config import load_config
+from TTS.config.shared_configs import BaseDatasetConfig
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.utils.managers import save_file
+from TTS.tts.utils.speakers import SpeakerManager
+
+parser = argparse.ArgumentParser(
+    description="""Compute embedding vectors for each audio file in a dataset and store them keyed by `{dataset_name}#{file_path}` in a .pth file\n\n"""
+    """
+    Example runs:
+    python TTS/bin/compute_embeddings.py --model_path speaker_encoder_model.pth --config_path speaker_encoder_config.json  --config_dataset_path dataset_config.json
+
+    python TTS/bin/compute_embeddings.py --model_path speaker_encoder_model.pth --config_path speaker_encoder_config.json  --fomatter vctk --dataset_path /path/to/vctk/dataset --dataset_name my_vctk --metafile /path/to/vctk/metafile.csv
+    """,
+    formatter_class=RawTextHelpFormatter,
+)
+parser.add_argument(
+    "--model_path",
+    type=str,
+    help="Path to model checkpoint file. It defaults to the released speaker encoder.",
+    default="https://github.com/coqui-ai/TTS/releases/download/speaker_encoder_model/model_se.pth.tar",
+)
+parser.add_argument(
+    "--config_path",
+    type=str,
+    help="Path to model config file. It defaults to the released speaker encoder config.",
+    default="https://github.com/coqui-ai/TTS/releases/download/speaker_encoder_model/config_se.json",
+)
+parser.add_argument(
+    "--config_dataset_path",
+    type=str,
+    help="Path to dataset config file. You either need to provide this or `formatter_name`, `dataset_name` and `dataset_path` arguments.",
+    default=None,
+)
+parser.add_argument("--output_path", type=str, help="Path for output `pth` or `json` file.", default="speakers.pth")
+parser.add_argument("--old_file", type=str, help="Previous embedding file to only compute new audios.", default=None)
+parser.add_argument("--disable_cuda", type=bool, help="Flag to disable cuda.", default=False)
+parser.add_argument("--no_eval", type=bool, help="Do not compute eval?. Default False", default=False)
+parser.add_argument(
+    "--formatter_name",
+    type=str,
+    help="Name of the formatter to use. You either need to provide this or `config_dataset_path`",
+    default=None,
+)
+parser.add_argument(
+    "--dataset_name",
+    type=str,
+    help="Name of the dataset to use. You either need to provide this or `config_dataset_path`",
+    default=None,
+)
+parser.add_argument(
+    "--dataset_path",
+    type=str,
+    help="Path to the dataset. You either need to provide this or `config_dataset_path`",
+    default=None,
+)
+parser.add_argument(
+    "--metafile",
+    type=str,
+    help="Path to the meta file. If not set, dataset formatter uses the default metafile if it is defined in the formatter. You either need to provide this or `config_dataset_path`",
+    default=None,
+)
+args = parser.parse_args()
+
+use_cuda = torch.cuda.is_available() and not args.disable_cuda
+
+if args.config_dataset_path is not None:
+    c_dataset = load_config(args.config_dataset_path)
+    meta_data_train, meta_data_eval = load_tts_samples(c_dataset.datasets, eval_split=not args.no_eval)
+else:
+    c_dataset = BaseDatasetConfig()
+    c_dataset.formatter = args.formatter_name
+    c_dataset.dataset_name = args.dataset_name
+    c_dataset.path = args.dataset_path
+    c_dataset.meta_file_train = args.metafile if args.metafile else None
+    meta_data_train, meta_data_eval = load_tts_samples(c_dataset, eval_split=not args.no_eval)
+
+
+if meta_data_eval is None:
+    samples = meta_data_train
+else:
+    samples = meta_data_train + meta_data_eval
+
+encoder_manager = SpeakerManager(
+    encoder_model_path=args.model_path,
+    encoder_config_path=args.config_path,
+    d_vectors_file_path=args.old_file,
+    use_cuda=use_cuda,
+)
+
+class_name_key = encoder_manager.encoder_config.class_name_key
+
+# compute speaker embeddings
+speaker_mapping = {}
+for idx, fields in enumerate(tqdm(samples)):
+    class_name = fields[class_name_key]
+    audio_file = fields["audio_file"]
+    embedding_key = fields["audio_unique_name"]
+    root_path = fields["root_path"]
+
+    if args.old_file is not None and embedding_key in encoder_manager.clip_ids:
+        # get the embedding from the old file
+        embedd = encoder_manager.get_embedding_by_clip(embedding_key)
+    else:
+        # extract the embedding
+        embedd = encoder_manager.compute_embedding_from_clip(audio_file)
+
+    # create speaker_mapping if target dataset is defined
+    speaker_mapping[embedding_key] = {}
+    speaker_mapping[embedding_key]["name"] = class_name
+    speaker_mapping[embedding_key]["embedding"] = embedd
+
+if speaker_mapping:
+    # save speaker_mapping if target dataset is defined
+    if os.path.isdir(args.output_path):
+        mapping_file_path = os.path.join(args.output_path, "speakers.pth")
+    else:
+        mapping_file_path = args.output_path
+
+    if os.path.dirname(mapping_file_path) != "":
+        os.makedirs(os.path.dirname(mapping_file_path), exist_ok=True)
+
+    save_file(speaker_mapping, mapping_file_path)
+    print("Speaker embeddings saved at:", mapping_file_path)

TTS/bin/compute_statistics.py

@@ -0,0 +1,98 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+import argparse
+import glob
+import os
+import sys
+
+sys.path.append('.')
+import numpy as np
+from tqdm import tqdm
+
+# from TTS.utils.io import load_config
+from TTS.config import load_config
+from TTS.tts.datasets import load_tts_samples
+from TTS.utils.audio import AudioProcessor
+
+
+def main():
+    """Run preprocessing process."""
+    parser = argparse.ArgumentParser(description="Compute mean and variance of spectrogtram features.")
+    parser.add_argument("config_path", type=str, help="TTS config file path to define audio processin parameters.")
+    parser.add_argument("out_path", type=str, help="save path (directory and filename).")
+    parser.add_argument(
+        "--data_path",
+        type=str,
+        required=False,
+        help="folder including the target set of wavs overriding dataset config.",
+    )
+    args, overrides = parser.parse_known_args()
+
+    CONFIG = load_config(args.config_path)
+    CONFIG.parse_known_args(overrides, relaxed_parser=True)
+
+    # load config
+    CONFIG.audio.signal_norm = False  # do not apply earlier normalization
+    CONFIG.audio.stats_path = None  # discard pre-defined stats
+
+    # load audio processor
+    ap = AudioProcessor(**CONFIG.audio.to_dict())
+
+    # load the meta data of target dataset
+    if args.data_path:
+        dataset_items = glob.glob(os.path.join(args.data_path, "**", "*.wav"), recursive=True)
+    else:
+        dataset_items = load_tts_samples(CONFIG.datasets)[0]  # take only train data
+    print(f" > There are {len(dataset_items)} files.")
+
+    mel_sum = 0
+    mel_square_sum = 0
+    linear_sum = 0
+    linear_square_sum = 0
+    N = 0
+    for item in tqdm(dataset_items):
+        # compute features
+        wav = ap.load_wav(item if isinstance(item, str) else item["audio_file"])
+        linear = ap.spectrogram(wav)
+        mel = ap.melspectrogram(wav)
+
+        # compute stats
+        N += mel.shape[1]
+        mel_sum += mel.sum(1)
+        linear_sum += linear.sum(1)
+        mel_square_sum += (mel**2).sum(axis=1)
+        linear_square_sum += (linear**2).sum(axis=1)
+
+    mel_mean = mel_sum / N
+    mel_scale = np.sqrt(mel_square_sum / N - mel_mean**2)
+    linear_mean = linear_sum / N
+    linear_scale = np.sqrt(linear_square_sum / N - linear_mean**2)
+
+    output_file_path = args.out_path
+    stats = {}
+    stats["mel_mean"] = mel_mean
+    stats["mel_std"] = mel_scale
+    stats["linear_mean"] = linear_mean
+    stats["linear_std"] = linear_scale
+
+    print(f" > Avg mel spec mean: {mel_mean.mean()}")
+    print(f" > Avg mel spec scale: {mel_scale.mean()}")
+    print(f" > Avg linear spec mean: {linear_mean.mean()}")
+    print(f" > Avg linear spec scale: {linear_scale.mean()}")
+
+    # set default config values for mean-var scaling
+    CONFIG.audio.stats_path = output_file_path
+    CONFIG.audio.signal_norm = True
+    # remove redundant values
+    del CONFIG.audio.max_norm
+    del CONFIG.audio.min_level_db
+    del CONFIG.audio.symmetric_norm
+    del CONFIG.audio.clip_norm
+    stats["audio_config"] = CONFIG.audio.to_dict()
+    np.save(output_file_path, stats, allow_pickle=True)
+    print(f" > stats saved to {output_file_path}")
+
+
+if __name__ == "__main__":
+    main()

TTS/bin/eval_encoder.py

@@ -0,0 +1,89 @@
+import argparse
+from argparse import RawTextHelpFormatter
+
+import torch
+from tqdm import tqdm
+
+from TTS.config import load_config
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.utils.speakers import SpeakerManager
+
+
+def compute_encoder_accuracy(dataset_items, encoder_manager):
+
+    class_name_key = encoder_manager.encoder_config.class_name_key
+    map_classid_to_classname = getattr(encoder_manager.encoder_config, "map_classid_to_classname", None)
+
+    class_acc_dict = {}
+
+    # compute embeddings for all wav_files
+    for item in tqdm(dataset_items):
+        class_name = item[class_name_key]
+        wav_file = item["audio_file"]
+
+        # extract the embedding
+        embedd = encoder_manager.compute_embedding_from_clip(wav_file)
+        if encoder_manager.encoder_criterion is not None and map_classid_to_classname is not None:
+            embedding = torch.FloatTensor(embedd).unsqueeze(0)
+            if encoder_manager.use_cuda:
+                embedding = embedding.cuda()
+
+            class_id = encoder_manager.encoder_criterion.softmax.inference(embedding).item()
+            predicted_label = map_classid_to_classname[str(class_id)]
+        else:
+            predicted_label = None
+
+        if class_name is not None and predicted_label is not None:
+            is_equal = int(class_name == predicted_label)
+            if class_name not in class_acc_dict:
+                class_acc_dict[class_name] = [is_equal]
+            else:
+                class_acc_dict[class_name].append(is_equal)
+        else:
+            raise RuntimeError("Error: class_name or/and predicted_label are None")
+
+    acc_avg = 0
+    for key, values in class_acc_dict.items():
+        acc = sum(values) / len(values)
+        print("Class", key, "Accuracy:", acc)
+        acc_avg += acc
+
+    print("Average Accuracy:", acc_avg / len(class_acc_dict))
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="""Compute the accuracy of the encoder.\n\n"""
+        """
+        Example runs:
+        python TTS/bin/eval_encoder.py emotion_encoder_model.pth emotion_encoder_config.json  dataset_config.json
+        """,
+        formatter_class=RawTextHelpFormatter,
+    )
+    parser.add_argument("model_path", type=str, help="Path to model checkpoint file.")
+    parser.add_argument(
+        "config_path",
+        type=str,
+        help="Path to model config file.",
+    )
+
+    parser.add_argument(
+        "config_dataset_path",
+        type=str,
+        help="Path to dataset config file.",
+    )
+    parser.add_argument("--use_cuda", type=bool, help="flag to set cuda.", default=True)
+    parser.add_argument("--eval", type=bool, help="compute eval.", default=True)
+
+    args = parser.parse_args()
+
+    c_dataset = load_config(args.config_dataset_path)
+
+    meta_data_train, meta_data_eval = load_tts_samples(c_dataset.datasets, eval_split=args.eval)
+    items = meta_data_train + meta_data_eval
+
+    enc_manager = SpeakerManager(
+        encoder_model_path=args.model_path, encoder_config_path=args.config_path, use_cuda=args.use_cuda
+    )
+
+    compute_encoder_accuracy(items, enc_manager)

TTS/bin/extract_tts_spectrograms.py

@@ -0,0 +1,287 @@
+#!/usr/bin/env python3
+"""Extract Mel spectrograms with teacher forcing."""
+
+import argparse
+import os
+
+import numpy as np
+import torch
+from torch.utils.data import DataLoader
+from tqdm import tqdm
+
+from TTS.config import load_config
+from TTS.tts.datasets import TTSDataset, load_tts_samples
+from TTS.tts.models import setup_model
+from TTS.tts.utils.speakers import SpeakerManager
+from TTS.tts.utils.text.tokenizer import TTSTokenizer
+from TTS.utils.audio import AudioProcessor
+from TTS.utils.generic_utils import count_parameters
+
+use_cuda = torch.cuda.is_available()
+
+
+def setup_loader(ap, r, verbose=False):
+    tokenizer, _ = TTSTokenizer.init_from_config(c)
+    dataset = TTSDataset(
+        outputs_per_step=r,
+        compute_linear_spec=False,
+        samples=meta_data,
+        tokenizer=tokenizer,
+        ap=ap,
+        batch_group_size=0,
+        min_text_len=c.min_text_len,
+        max_text_len=c.max_text_len,
+        min_audio_len=c.min_audio_len,
+        max_audio_len=c.max_audio_len,
+        phoneme_cache_path=c.phoneme_cache_path,
+        precompute_num_workers=0,
+        use_noise_augment=False,
+        verbose=verbose,
+        speaker_id_mapping=speaker_manager.name_to_id if c.use_speaker_embedding else None,
+        d_vector_mapping=speaker_manager.embeddings if c.use_d_vector_file else None,
+    )
+
+    if c.use_phonemes and c.compute_input_seq_cache:
+        # precompute phonemes to have a better estimate of sequence lengths.
+        dataset.compute_input_seq(c.num_loader_workers)
+    dataset.preprocess_samples()
+
+    loader = DataLoader(
+        dataset,
+        batch_size=c.batch_size,
+        shuffle=False,
+        collate_fn=dataset.collate_fn,
+        drop_last=False,
+        sampler=None,
+        num_workers=c.num_loader_workers,
+        pin_memory=False,
+    )
+    return loader
+
+
+def set_filename(wav_path, out_path):
+    wav_file = os.path.basename(wav_path)
+    file_name = wav_file.split(".")[0]
+    os.makedirs(os.path.join(out_path, "quant"), exist_ok=True)
+    os.makedirs(os.path.join(out_path, "mel"), exist_ok=True)
+    os.makedirs(os.path.join(out_path, "wav_gl"), exist_ok=True)
+    os.makedirs(os.path.join(out_path, "wav"), exist_ok=True)
+    wavq_path = os.path.join(out_path, "quant", file_name)
+    mel_path = os.path.join(out_path, "mel", file_name)
+    wav_gl_path = os.path.join(out_path, "wav_gl", file_name + ".wav")
+    wav_path = os.path.join(out_path, "wav", file_name + ".wav")
+    return file_name, wavq_path, mel_path, wav_gl_path, wav_path
+
+
+def format_data(data):
+    # setup input data
+    text_input = data["token_id"]
+    text_lengths = data["token_id_lengths"]
+    mel_input = data["mel"]
+    mel_lengths = data["mel_lengths"]
+    item_idx = data["item_idxs"]
+    d_vectors = data["d_vectors"]
+    speaker_ids = data["speaker_ids"]
+    attn_mask = data["attns"]
+    avg_text_length = torch.mean(text_lengths.float())
+    avg_spec_length = torch.mean(mel_lengths.float())
+
+    # dispatch data to GPU
+    if use_cuda:
+        text_input = text_input.cuda(non_blocking=True)
+        text_lengths = text_lengths.cuda(non_blocking=True)
+        mel_input = mel_input.cuda(non_blocking=True)
+        mel_lengths = mel_lengths.cuda(non_blocking=True)
+        if speaker_ids is not None:
+            speaker_ids = speaker_ids.cuda(non_blocking=True)
+        if d_vectors is not None:
+            d_vectors = d_vectors.cuda(non_blocking=True)
+        if attn_mask is not None:
+            attn_mask = attn_mask.cuda(non_blocking=True)
+    return (
+        text_input,
+        text_lengths,
+        mel_input,
+        mel_lengths,
+        speaker_ids,
+        d_vectors,
+        avg_text_length,
+        avg_spec_length,
+        attn_mask,
+        item_idx,
+    )
+
+
+@torch.no_grad()
+def inference(
+    model_name,
+    model,
+    ap,
+    text_input,
+    text_lengths,
+    mel_input,
+    mel_lengths,
+    speaker_ids=None,
+    d_vectors=None,
+):
+    if model_name == "glow_tts":
+        speaker_c = None
+        if speaker_ids is not None:
+            speaker_c = speaker_ids
+        elif d_vectors is not None:
+            speaker_c = d_vectors
+        outputs = model.inference_with_MAS(
+            text_input,
+            text_lengths,
+            mel_input,
+            mel_lengths,
+            aux_input={"d_vectors": speaker_c, "speaker_ids": speaker_ids},
+        )
+        model_output = outputs["model_outputs"]
+        model_output = model_output.detach().cpu().numpy()
+
+    elif "tacotron" in model_name:
+        aux_input = {"speaker_ids": speaker_ids, "d_vectors": d_vectors}
+        outputs = model(text_input, text_lengths, mel_input, mel_lengths, aux_input)
+        postnet_outputs = outputs["model_outputs"]
+        # normalize tacotron output
+        if model_name == "tacotron":
+            mel_specs = []
+            postnet_outputs = postnet_outputs.data.cpu().numpy()
+            for b in range(postnet_outputs.shape[0]):
+                postnet_output = postnet_outputs[b]
+                mel_specs.append(torch.FloatTensor(ap.out_linear_to_mel(postnet_output.T).T))
+            model_output = torch.stack(mel_specs).cpu().numpy()
+
+        elif model_name == "tacotron2":
+            model_output = postnet_outputs.detach().cpu().numpy()
+    return model_output
+
+
+def extract_spectrograms(
+    data_loader, model, ap, output_path, quantized_wav=False, save_audio=False, debug=False, metada_name="metada.txt"
+):
+    model.eval()
+    export_metadata = []
+    for _, data in tqdm(enumerate(data_loader), total=len(data_loader)):
+
+        # format data
+        (
+            text_input,
+            text_lengths,
+            mel_input,
+            mel_lengths,
+            speaker_ids,
+            d_vectors,
+            _,
+            _,
+            _,
+            item_idx,
+        ) = format_data(data)
+
+        model_output = inference(
+            c.model.lower(),
+            model,
+            ap,
+            text_input,
+            text_lengths,
+            mel_input,
+            mel_lengths,
+            speaker_ids,
+            d_vectors,
+        )
+
+        for idx in range(text_input.shape[0]):
+            wav_file_path = item_idx[idx]
+            wav = ap.load_wav(wav_file_path)
+            _, wavq_path, mel_path, wav_gl_path, wav_path = set_filename(wav_file_path, output_path)
+
+            # quantize and save wav
+            if quantized_wav:
+                wavq = ap.quantize(wav)
+                np.save(wavq_path, wavq)
+
+            # save TTS mel
+            mel = model_output[idx]
+            mel_length = mel_lengths[idx]
+            mel = mel[:mel_length, :].T
+            np.save(mel_path, mel)
+
+            export_metadata.append([wav_file_path, mel_path])
+            if save_audio:
+                ap.save_wav(wav, wav_path)
+
+            if debug:
+                print("Audio for debug saved at:", wav_gl_path)
+                wav = ap.inv_melspectrogram(mel)
+                ap.save_wav(wav, wav_gl_path)
+
+    with open(os.path.join(output_path, metada_name), "w", encoding="utf-8") as f:
+        for data in export_metadata:
+            f.write(f"{data[0]}|{data[1]+'.npy'}\n")
+
+
+def main(args):  # pylint: disable=redefined-outer-name
+    # pylint: disable=global-variable-undefined
+    global meta_data, speaker_manager
+
+    # Audio processor
+    ap = AudioProcessor(**c.audio)
+
+    # load data instances
+    meta_data_train, meta_data_eval = load_tts_samples(
+        c.datasets, eval_split=args.eval, eval_split_max_size=c.eval_split_max_size, eval_split_size=c.eval_split_size
+    )
+
+    # use eval and training partitions
+    meta_data = meta_data_train + meta_data_eval
+
+    # init speaker manager
+    if c.use_speaker_embedding:
+        speaker_manager = SpeakerManager(data_items=meta_data)
+    elif c.use_d_vector_file:
+        speaker_manager = SpeakerManager(d_vectors_file_path=c.d_vector_file)
+    else:
+        speaker_manager = None
+
+    # setup model
+    model = setup_model(c)
+
+    # restore model
+    model.load_checkpoint(c, args.checkpoint_path, eval=True)
+
+    if use_cuda:
+        model.cuda()
+
+    num_params = count_parameters(model)
+    print("\n > Model has {} parameters".format(num_params), flush=True)
+    # set r
+    r = 1 if c.model.lower() == "glow_tts" else model.decoder.r
+    own_loader = setup_loader(ap, r, verbose=True)
+
+    extract_spectrograms(
+        own_loader,
+        model,
+        ap,
+        args.output_path,
+        quantized_wav=args.quantized,
+        save_audio=args.save_audio,
+        debug=args.debug,
+        metada_name="metada.txt",
+    )
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--config_path", type=str, help="Path to config file for training.", required=True)
+    parser.add_argument("--checkpoint_path", type=str, help="Model file to be restored.", required=True)
+    parser.add_argument("--output_path", type=str, help="Path to save mel specs", required=True)
+    parser.add_argument("--debug", default=False, action="store_true", help="Save audio files for debug")
+    parser.add_argument("--save_audio", default=False, action="store_true", help="Save audio files")
+    parser.add_argument("--quantized", action="store_true", help="Save quantized audio files")
+    parser.add_argument("--eval", type=bool, help="compute eval.", default=True)
+    args = parser.parse_args()
+
+    c = load_config(args.config_path)
+    c.audio.trim_silence = False
+    main(args)

TTS/bin/find_unique_chars.py

@@ -0,0 +1,48 @@
+"""Find all the unique characters in a dataset"""
+import argparse
+from argparse import RawTextHelpFormatter
+
+import sys
+
+sys.path.append('.')
+from TTS.config import load_config
+from TTS.tts.datasets import load_tts_samples
+
+
+def main():
+    # pylint: disable=bad-option-value
+    parser = argparse.ArgumentParser(
+        description="""Find all the unique characters or phonemes in a dataset.\n\n"""
+        """
+    Example runs:
+
+    python TTS/bin/find_unique_chars.py --config_path config.json
+    """,
+        formatter_class=RawTextHelpFormatter,
+    )
+    parser.add_argument("--config_path", type=str, help="Path to dataset config file.", required=True)
+    args = parser.parse_args()
+
+    c = load_config(args.config_path)
+
+    # load all datasets
+    train_items, eval_items = load_tts_samples(
+        c.datasets, eval_split=True, eval_split_max_size=c.eval_split_max_size, eval_split_size=c.eval_split_size
+    )
+
+    items = train_items + eval_items
+
+    texts = "".join(item["text"] for item in items)
+    chars = set(texts)
+    lower_chars = filter(lambda c: c.islower(), chars)
+    chars_force_lower = [c.lower() for c in chars]
+    chars_force_lower = set(chars_force_lower)
+
+    print(f" > Number of unique characters: {len(chars)}")
+    print(f" > Unique characters: {''.join(sorted(chars))}")
+    print(f" > Unique lower characters: {''.join(sorted(lower_chars))}")
+    print(f" > Unique all forced to lower characters: {''.join(sorted(chars_force_lower))}")
+
+
+if __name__ == "__main__":
+    main()

TTS/bin/find_unique_phonemes.py

@@ -0,0 +1,78 @@
+"""Find all the unique characters in a dataset"""
+import argparse
+import multiprocessing
+import sys
+
+sys.path.append('.')
+from argparse import RawTextHelpFormatter
+
+from tqdm.contrib.concurrent import process_map
+
+from TTS.config import load_config
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.utils.text.phonemizers import Gruut,ESpeak
+
+
+def compute_phonemes(item):
+    text = item["text"]
+    ph = phonemizer.phonemize(text).replace("|", "")
+    return set(list(ph))
+
+
+def main():
+    # pylint: disable=W0601
+    global c, phonemizer
+    # pylint: disable=bad-option-value
+    parser = argparse.ArgumentParser(
+        description="""Find all the unique characters or phonemes in a dataset.\n\n"""
+        """
+    Example runs:
+
+    python TTS/bin/find_unique_phonemes.py --config_path config.json
+    """,
+        formatter_class=RawTextHelpFormatter,
+    )
+    parser.add_argument("--config_path", type=str, help="Path to dataset config file.", required=True)
+    args = parser.parse_args()
+
+    c = load_config(args.config_path)
+
+    # load all datasets
+    train_items, eval_items = load_tts_samples(
+        c.datasets, eval_split=True, eval_split_max_size=c.eval_split_max_size, eval_split_size=c.eval_split_size
+    )
+    items = train_items + eval_items
+    print("Num items:", len(items))
+
+    language_list = [item["language"] for item in items]
+    is_lang_def = all(language_list)
+
+    if not c.phoneme_language or not is_lang_def:
+        raise ValueError("Phoneme language must be defined in config.")
+
+    if not language_list.count(language_list[0]) == len(language_list):
+        raise ValueError(
+            "Currently, just one phoneme language per config file is supported !! Please split the dataset config into different configs and run it individually for each language !!"
+        )
+
+    # phonemizer = Gruut(language=language_list[0], keep_puncs=True)
+    phonemizer = ESpeak(language="vi", backend="espeak")
+    phonemes = process_map(compute_phonemes, items, max_workers=multiprocessing.cpu_count(), chunksize=15)
+    print(phonemes)
+    phones = []
+    for ph in phonemes:
+        phones.extend(ph)
+
+    phones = set(phones)
+    lower_phones = filter(lambda c: c.islower(), phones)
+    phones_force_lower = [c.lower() for c in phones]
+    phones_force_lower = set(phones_force_lower)
+
+    print(f" > Number of unique phonemes: {len(phones)}")
+    print(f" > Unique phonemes: {''.join(sorted(phones))}")
+    print(f" > Unique lower phonemes: {''.join(sorted(lower_phones))}")
+    print(f" > Unique all forced to lower phonemes: {''.join(sorted(phones_force_lower))}")
+
+
+if __name__ == "__main__":
+    main()

TTS/bin/remove_silence_using_vad.py

@@ -0,0 +1,93 @@
+import argparse
+import glob
+import os
+import pathlib
+
+from tqdm import tqdm
+
+from TTS.utils.vad import get_vad_model_and_utils, remove_silence
+
+
+def adjust_path_and_remove_silence(audio_path):
+    output_path = audio_path.replace(os.path.join(args.input_dir, ""), os.path.join(args.output_dir, ""))
+    # ignore if the file exists
+    if os.path.exists(output_path) and not args.force:
+        return output_path
+
+    # create all directory structure
+    pathlib.Path(output_path).parent.mkdir(parents=True, exist_ok=True)
+    # remove the silence and save the audio
+    output_path, is_speech = remove_silence(
+        model_and_utils,
+        audio_path,
+        output_path,
+        trim_just_beginning_and_end=args.trim_just_beginning_and_end,
+        use_cuda=args.use_cuda,
+    )
+
+    return output_path, is_speech
+
+
+def preprocess_audios():
+    files = sorted(glob.glob(os.path.join(args.input_dir, args.glob), recursive=True))
+    print("> Number of files: ", len(files))
+    if not args.force:
+        print("> Ignoring files that already exist in the output idrectory.")
+
+    if args.trim_just_beginning_and_end:
+        print("> Trimming just the beginning and the end with nonspeech parts.")
+    else:
+        print("> Trimming all nonspeech parts.")
+
+    filtered_files = []
+    if files:
+        # create threads
+        # num_threads = multiprocessing.cpu_count()
+        # process_map(adjust_path_and_remove_silence, files, max_workers=num_threads, chunksize=15)
+        for f in tqdm(files):
+            output_path, is_speech = adjust_path_and_remove_silence(f)
+            if not is_speech:
+                filtered_files.append(output_path)
+
+        # write files that do not have speech
+        with open(os.path.join(args.output_dir, "filtered_files.txt"), "w", encoding="utf-8") as f:
+            for file in filtered_files:
+                f.write(file + "\n")
+    else:
+        print("> No files Found !")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="python TTS/bin/remove_silence_using_vad.py -i=VCTK-Corpus/ -o=VCTK-Corpus-removed-silence/ -g=wav48_silence_trimmed/*/*_mic1.flac --trim_just_beginning_and_end True"
+    )
+    parser.add_argument("-i", "--input_dir", type=str, default="../VCTK-Corpus", help="Dataset root dir")
+    parser.add_argument(
+        "-o", "--output_dir", type=str, default="../VCTK-Corpus-removed-silence", help="Output Dataset dir"
+    )
+    parser.add_argument("-f", "--force", default=False, action="store_true", help="Force the replace of exists files")
+    parser.add_argument(
+        "-g",
+        "--glob",
+        type=str,
+        default="**/*.wav",
+        help="path in glob format for acess wavs from input_dir. ex: wav48/*/*.wav",
+    )
+    parser.add_argument(
+        "-t",
+        "--trim_just_beginning_and_end",
+        type=bool,
+        default=True,
+        help="If True this script will trim just the beginning and end nonspeech parts. If False all nonspeech parts will be trim. Default True",
+    )
+    parser.add_argument(
+        "-c",
+        "--use_cuda",
+        type=bool,
+        default=False,
+        help="If True use cuda",
+    )
+    args = parser.parse_args()
+    # load the model and utils
+    model_and_utils = get_vad_model_and_utils(use_cuda=args.use_cuda)
+    preprocess_audios()

TTS/bin/resample.py

@@ -0,0 +1,87 @@
+import argparse
+import glob
+import os
+from argparse import RawTextHelpFormatter
+from distutils.dir_util import copy_tree
+from multiprocessing import Pool
+
+import librosa
+import soundfile as sf
+from tqdm import tqdm
+
+
+def resample_file(func_args):
+    filename, output_sr = func_args
+    y, sr = librosa.load(filename, sr=output_sr)
+    sf.write(filename, y, sr)
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser(
+        description="""Resample a folder recusively with librosa
+                       Can be used in place or create a copy of the folder as an output.\n\n
+                       Example run:
+                            python TTS/bin/resample.py
+                                --input_dir /root/LJSpeech-1.1/
+                                --output_sr 22050
+                                --output_dir /root/resampled_LJSpeech-1.1/
+                                --file_ext wav
+                                --n_jobs 24
+                    """,
+        formatter_class=RawTextHelpFormatter,
+    )
+
+    parser.add_argument(
+        "--input_dir",
+        type=str,
+        default=None,
+        required=True,
+        help="Path of the folder containing the audio files to resample",
+    )
+
+    parser.add_argument(
+        "--output_sr",
+        type=int,
+        default=22050,
+        required=False,
+        help="Samlple rate to which the audio files should be resampled",
+    )
+
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default=None,
+        required=False,
+        help="Path of the destination folder. If not defined, the operation is done in place",
+    )
+
+    parser.add_argument(
+        "--file_ext",
+        type=str,
+        default="wav",
+        required=False,
+        help="Extension of the audio files to resample",
+    )
+
+    parser.add_argument(
+        "--n_jobs", type=int, default=None, help="Number of threads to use, by default it uses all cores"
+    )
+
+    args = parser.parse_args()
+
+    if args.output_dir:
+        print("Recursively copying the input folder...")
+        copy_tree(args.input_dir, args.output_dir)
+        args.input_dir = args.output_dir
+
+    print("Resampling the audio files...")
+    audio_files = glob.glob(os.path.join(args.input_dir, f"**/*.{args.file_ext}"), recursive=True)
+    print(f"Found {len(audio_files)} files...")
+    audio_files = list(zip(audio_files, len(audio_files) * [args.output_sr]))
+    with Pool(processes=args.n_jobs) as p:
+        with tqdm(total=len(audio_files)) as pbar:
+            for i, _ in enumerate(p.imap_unordered(resample_file, audio_files)):
+                pbar.update()
+
+    print("Done !")

TTS/bin/synthesize.py

@@ -0,0 +1,374 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+import argparse
+import sys
+from argparse import RawTextHelpFormatter
+
+# pylint: disable=redefined-outer-name, unused-argument
+from pathlib import Path
+
+from TTS.utils.manage import ModelManager
+from TTS.utils.synthesizer import Synthesizer
+
+
+def str2bool(v):
+    if isinstance(v, bool):
+        return v
+    if v.lower() in ("yes", "true", "t", "y", "1"):
+        return True
+    if v.lower() in ("no", "false", "f", "n", "0"):
+        return False
+    raise argparse.ArgumentTypeError("Boolean value expected.")
+
+
+def main():
+    description = """Synthesize speech on command line.
+
+You can either use your trained model or choose a model from the provided list.
+
+If you don't specify any models, then it uses LJSpeech based English model.
+
+## Example Runs
+
+### Single Speaker Models
+
+- List provided models:
+
+    ```
+    $ tts --list_models
+    ```
+
+- Query info for model info by idx:
+
+    ```
+    $ tts --model_info_by_idx "<model_type>/<model_query_idx>"
+    ```
+
+- Query info for model info by full name:
+
+    ```
+    $ tts --model_info_by_name "<model_type>/<language>/<dataset>/<model_name>"
+    ```
+
+- Run TTS with default models:
+
+    ```
+    $ tts --text "Text for TTS"
+    ```
+
+- Run a TTS model with its default vocoder model:
+
+    ```
+    $ tts --text "Text for TTS" --model_name "<model_type>/<language>/<dataset>/<model_name>
+    ```
+
+- Run with specific TTS and vocoder models from the list:
+
+    ```
+    $ tts --text "Text for TTS" --model_name "<model_type>/<language>/<dataset>/<model_name>" --vocoder_name "<model_type>/<language>/<dataset>/<model_name>" --output_path
+    ```
+
+- Run your own TTS model (Using Griffin-Lim Vocoder):
+
+    ```
+    $ tts --text "Text for TTS" --model_path path/to/model.pth --config_path path/to/config.json --out_path output/path/speech.wav
+    ```
+
+- Run your own TTS and Vocoder models:
+    ```
+    $ tts --text "Text for TTS" --model_path path/to/config.json --config_path path/to/model.pth --out_path output/path/speech.wav
+        --vocoder_path path/to/vocoder.pth --vocoder_config_path path/to/vocoder_config.json
+    ```
+
+### Multi-speaker Models
+
+- List the available speakers and choose as <speaker_id> among them:
+
+    ```
+    $ tts --model_name "<language>/<dataset>/<model_name>"  --list_speaker_idxs
+    ```
+
+- Run the multi-speaker TTS model with the target speaker ID:
+
+    ```
+    $ tts --text "Text for TTS." --out_path output/path/speech.wav --model_name "<language>/<dataset>/<model_name>"  --speaker_idx <speaker_id>
+    ```
+
+- Run your own multi-speaker TTS model:
+
+    ```
+    $ tts --text "Text for TTS" --out_path output/path/speech.wav --model_path path/to/config.json --config_path path/to/model.pth --speakers_file_path path/to/speaker.json --speaker_idx <speaker_id>
+    ```
+    """
+    # We remove Markdown code formatting programmatically here to allow us to copy-and-paste from main README to keep
+    # documentation in sync more easily.
+    parser = argparse.ArgumentParser(
+        description=description.replace("    ```\n", ""),
+        formatter_class=RawTextHelpFormatter,
+    )
+
+    parser.add_argument(
+        "--list_models",
+        type=str2bool,
+        nargs="?",
+        const=True,
+        default=False,
+        help="list available pre-trained TTS and vocoder models.",
+    )
+
+    parser.add_argument(
+        "--model_info_by_idx",
+        type=str,
+        default=None,
+        help="model info using query format: <model_type>/<model_query_idx>",
+    )
+
+    parser.add_argument(
+        "--model_info_by_name",
+        type=str,
+        default=None,
+        help="model info using query format: <model_type>/<language>/<dataset>/<model_name>",
+    )
+
+    parser.add_argument("--text", type=str, default=None, help="Text to generate speech.")
+
+    # Args for running pre-trained TTS models.
+    parser.add_argument(
+        "--model_name",
+        type=str,
+        default="tts_models/en/ljspeech/tacotron2-DDC",
+        help="Name of one of the pre-trained TTS models in format <language>/<dataset>/<model_name>",
+    )
+    parser.add_argument(
+        "--vocoder_name",
+        type=str,
+        default=None,
+        help="Name of one of the pre-trained  vocoder models in format <language>/<dataset>/<model_name>",
+    )
+
+    # Args for running custom models
+    parser.add_argument("--config_path", default=None, type=str, help="Path to model config file.")
+    parser.add_argument(
+        "--model_path",
+        type=str,
+        default=None,
+        help="Path to model file.",
+    )
+    parser.add_argument(
+        "--out_path",
+        type=str,
+        default="tts_output.wav",
+        help="Output wav file path.",
+    )
+    parser.add_argument("--use_cuda", type=bool, help="Run model on CUDA.", default=False)
+    parser.add_argument(
+        "--vocoder_path",
+        type=str,
+        help="Path to vocoder model file. If it is not defined, model uses GL as vocoder. Please make sure that you installed vocoder library before (WaveRNN).",
+        default=None,
+    )
+    parser.add_argument("--vocoder_config_path", type=str, help="Path to vocoder model config file.", default=None)
+    parser.add_argument(
+        "--encoder_path",
+        type=str,
+        help="Path to speaker encoder model file.",
+        default=None,
+    )
+    parser.add_argument("--encoder_config_path", type=str, help="Path to speaker encoder config file.", default=None)
+
+    # args for multi-speaker synthesis
+    parser.add_argument("--speakers_file_path", type=str, help="JSON file for multi-speaker model.", default=None)
+    parser.add_argument("--language_ids_file_path", type=str, help="JSON file for multi-lingual model.", default=None)
+    parser.add_argument(
+        "--speaker_idx",
+        type=str,
+        help="Target speaker ID for a multi-speaker TTS model.",
+        default=None,
+    )
+    parser.add_argument(
+        "--language_idx",
+        type=str,
+        help="Target language ID for a multi-lingual TTS model.",
+        default=None,
+    )
+    parser.add_argument(
+        "--speaker_wav",
+        nargs="+",
+        help="wav file(s) to condition a multi-speaker TTS model with a Speaker Encoder. You can give multiple file paths. The d_vectors is computed as their average.",
+        default=None,
+    )
+    parser.add_argument("--gst_style", help="Wav path file for GST style reference.", default=None)
+    parser.add_argument(
+        "--capacitron_style_wav", type=str, help="Wav path file for Capacitron prosody reference.", default=None
+    )
+    parser.add_argument("--capacitron_style_text", type=str, help="Transcription of the reference.", default=None)
+    parser.add_argument(
+        "--list_speaker_idxs",
+        help="List available speaker ids for the defined multi-speaker model.",
+        type=str2bool,
+        nargs="?",
+        const=True,
+        default=False,
+    )
+    parser.add_argument(
+        "--list_language_idxs",
+        help="List available language ids for the defined multi-lingual model.",
+        type=str2bool,
+        nargs="?",
+        const=True,
+        default=False,
+    )
+    # aux args
+    parser.add_argument(
+        "--save_spectogram",
+        type=bool,
+        help="If true save raw spectogram for further (vocoder) processing in out_path.",
+        default=False,
+    )
+    parser.add_argument(
+        "--reference_wav",
+        type=str,
+        help="Reference wav file to convert in the voice of the speaker_idx or speaker_wav",
+        default=None,
+    )
+    parser.add_argument(
+        "--reference_speaker_idx",
+        type=str,
+        help="speaker ID of the reference_wav speaker (If not provided the embedding will be computed using the Speaker Encoder).",
+        default=None,
+    )
+    parser.add_argument(
+        "--progress_bar",
+        type=str2bool,
+        help="If true shows a progress bar for the model download. Defaults to True",
+        default=True,
+    )
+
+    args = parser.parse_args()
+
+    # print the description if either text or list_models is not set
+    check_args = [
+        args.text,
+        args.list_models,
+        args.list_speaker_idxs,
+        args.list_language_idxs,
+        args.reference_wav,
+        args.model_info_by_idx,
+        args.model_info_by_name,
+    ]
+    if not any(check_args):
+        parser.parse_args(["-h"])
+
+    # load model manager
+    path = Path(__file__).parent / "../.models.json"
+    manager = ModelManager(path, progress_bar=args.progress_bar)
+
+    model_path = None
+    config_path = None
+    speakers_file_path = None
+    language_ids_file_path = None
+    vocoder_path = None
+    vocoder_config_path = None
+    encoder_path = None
+    encoder_config_path = None
+
+    # CASE1 #list : list pre-trained TTS models
+    if args.list_models:
+        manager.list_models()
+        sys.exit()
+
+    # CASE2 #info : model info of pre-trained TTS models
+    if args.model_info_by_idx:
+        model_query = args.model_info_by_idx
+        manager.model_info_by_idx(model_query)
+        sys.exit()
+
+    if args.model_info_by_name:
+        model_query_full_name = args.model_info_by_name
+        manager.model_info_by_full_name(model_query_full_name)
+        sys.exit()
+
+    # CASE3: load pre-trained model paths
+    if args.model_name is not None and not args.model_path:
+        model_path, config_path, model_item = manager.download_model(args.model_name)
+        args.vocoder_name = model_item["default_vocoder"] if args.vocoder_name is None else args.vocoder_name
+
+    if args.vocoder_name is not None and not args.vocoder_path:
+        vocoder_path, vocoder_config_path, _ = manager.download_model(args.vocoder_name)
+
+    # CASE4: set custom model paths
+    if args.model_path is not None:
+        model_path = args.model_path
+        config_path = args.config_path
+        speakers_file_path = args.speakers_file_path
+        language_ids_file_path = args.language_ids_file_path
+
+    if args.vocoder_path is not None:
+        vocoder_path = args.vocoder_path
+        vocoder_config_path = args.vocoder_config_path
+
+    if args.encoder_path is not None:
+        encoder_path = args.encoder_path
+        encoder_config_path = args.encoder_config_path
+
+    # load models
+    synthesizer = Synthesizer(
+        model_path,
+        config_path,
+        speakers_file_path,
+        language_ids_file_path,
+        vocoder_path,
+        vocoder_config_path,
+        encoder_path,
+        encoder_config_path,
+        args.use_cuda,
+    )
+
+    # query speaker ids of a multi-speaker model.
+    if args.list_speaker_idxs:
+        print(
+            " > Available speaker ids: (Set --speaker_idx flag to one of these values to use the multi-speaker model."
+        )
+        print(synthesizer.tts_model.speaker_manager.name_to_id)
+        return
+
+    # query langauge ids of a multi-lingual model.
+    if args.list_language_idxs:
+        print(
+            " > Available language ids: (Set --language_idx flag to one of these values to use the multi-lingual model."
+        )
+        print(synthesizer.tts_model.language_manager.name_to_id)
+        return
+
+    # check the arguments against a multi-speaker model.
+    if synthesizer.tts_speakers_file and (not args.speaker_idx and not args.speaker_wav):
+        print(
+            " [!] Looks like you use a multi-speaker model. Define `--speaker_idx` to "
+            "select the target speaker. You can list the available speakers for this model by `--list_speaker_idxs`."
+        )
+        return
+
+    # RUN THE SYNTHESIS
+    if args.text:
+        print(" > Text: {}".format(args.text))
+
+    # kick it
+    wav = synthesizer.tts(
+        args.text,
+        args.speaker_idx,
+        args.language_idx,
+        args.speaker_wav,
+        reference_wav=args.reference_wav,
+        style_wav=args.capacitron_style_wav,
+        style_text=args.capacitron_style_text,
+        reference_speaker_name=args.reference_speaker_idx,
+    )
+
+    # save the results
+    print(" > Saving output to {}".format(args.out_path))
+    synthesizer.save_wav(wav, args.out_path)
+
+
+if __name__ == "__main__":
+    main()

TTS/bin/train_encoder.py

@@ -0,0 +1,319 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+import os
+import sys
+import time
+import traceback
+
+import torch
+from torch.utils.data import DataLoader
+from trainer.torch import NoamLR
+from trainer.trainer_utils import get_optimizer
+
+from TTS.encoder.dataset import EncoderDataset
+from TTS.encoder.utils.generic_utils import save_best_model, save_checkpoint, setup_encoder_model
+from TTS.encoder.utils.training import init_training
+from TTS.encoder.utils.visual import plot_embeddings
+from TTS.tts.datasets import load_tts_samples
+from TTS.utils.audio import AudioProcessor
+from TTS.utils.generic_utils import count_parameters, remove_experiment_folder
+from TTS.utils.io import copy_model_files
+from TTS.utils.samplers import PerfectBatchSampler
+from TTS.utils.training import check_update
+
+torch.backends.cudnn.enabled = True
+torch.backends.cudnn.benchmark = True
+torch.manual_seed(54321)
+use_cuda = torch.cuda.is_available()
+num_gpus = torch.cuda.device_count()
+print(" > Using CUDA: ", use_cuda)
+print(" > Number of GPUs: ", num_gpus)
+
+
+def setup_loader(ap: AudioProcessor, is_val: bool = False, verbose: bool = False):
+    num_utter_per_class = c.num_utter_per_class if not is_val else c.eval_num_utter_per_class
+    num_classes_in_batch = c.num_classes_in_batch if not is_val else c.eval_num_classes_in_batch
+
+    dataset = EncoderDataset(
+        c,
+        ap,
+        meta_data_eval if is_val else meta_data_train,
+        voice_len=c.voice_len,
+        num_utter_per_class=num_utter_per_class,
+        num_classes_in_batch=num_classes_in_batch,
+        verbose=verbose,
+        augmentation_config=c.audio_augmentation if not is_val else None,
+        use_torch_spec=c.model_params.get("use_torch_spec", False),
+    )
+    # get classes list
+    classes = dataset.get_class_list()
+
+    sampler = PerfectBatchSampler(
+        dataset.items,
+        classes,
+        batch_size=num_classes_in_batch * num_utter_per_class,  # total batch size
+        num_classes_in_batch=num_classes_in_batch,
+        num_gpus=1,
+        shuffle=not is_val,
+        drop_last=True,
+    )
+
+    if len(classes) < num_classes_in_batch:
+        if is_val:
+            raise RuntimeError(
+                f"config.eval_num_classes_in_batch ({num_classes_in_batch}) need to be <= {len(classes)} (Number total of Classes in the Eval dataset) !"
+            )
+        raise RuntimeError(
+            f"config.num_classes_in_batch ({num_classes_in_batch}) need to be <= {len(classes)} (Number total of Classes in the Train dataset) !"
+        )
+
+    # set the classes to avoid get wrong class_id when the number of training and eval classes are not equal
+    if is_val:
+        dataset.set_classes(train_classes)
+
+    loader = DataLoader(
+        dataset,
+        num_workers=c.num_loader_workers,
+        batch_sampler=sampler,
+        collate_fn=dataset.collate_fn,
+    )
+
+    return loader, classes, dataset.get_map_classid_to_classname()
+
+
+def evaluation(model, criterion, data_loader, global_step):
+    eval_loss = 0
+    for _, data in enumerate(data_loader):
+        with torch.no_grad():
+            # setup input data
+            inputs, labels = data
+
+            # agroup samples of each class in the batch. perfect sampler produces [3,2,1,3,2,1] we need [3,3,2,2,1,1]
+            labels = torch.transpose(
+                labels.view(c.eval_num_utter_per_class, c.eval_num_classes_in_batch), 0, 1
+            ).reshape(labels.shape)
+            inputs = torch.transpose(
+                inputs.view(c.eval_num_utter_per_class, c.eval_num_classes_in_batch, -1), 0, 1
+            ).reshape(inputs.shape)
+
+            # dispatch data to GPU
+            if use_cuda:
+                inputs = inputs.cuda(non_blocking=True)
+                labels = labels.cuda(non_blocking=True)
+
+            # forward pass model
+            outputs = model(inputs)
+
+            # loss computation
+            loss = criterion(
+                outputs.view(c.eval_num_classes_in_batch, outputs.shape[0] // c.eval_num_classes_in_batch, -1), labels
+            )
+
+            eval_loss += loss.item()
+
+    eval_avg_loss = eval_loss / len(data_loader)
+    # save stats
+    dashboard_logger.eval_stats(global_step, {"loss": eval_avg_loss})
+    # plot the last batch in the evaluation
+    figures = {
+        "UMAP Plot": plot_embeddings(outputs.detach().cpu().numpy(), c.num_classes_in_batch),
+    }
+    dashboard_logger.eval_figures(global_step, figures)
+    return eval_avg_loss
+
+
+def train(model, optimizer, scheduler, criterion, data_loader, eval_data_loader, global_step):
+    model.train()
+    best_loss = float("inf")
+    avg_loader_time = 0
+    end_time = time.time()
+    for epoch in range(c.epochs):
+        tot_loss = 0
+        epoch_time = 0
+        for _, data in enumerate(data_loader):
+            start_time = time.time()
+
+            # setup input data
+            inputs, labels = data
+            # agroup samples of each class in the batch. perfect sampler produces [3,2,1,3,2,1] we need [3,3,2,2,1,1]
+            labels = torch.transpose(labels.view(c.num_utter_per_class, c.num_classes_in_batch), 0, 1).reshape(
+                labels.shape
+            )
+            inputs = torch.transpose(inputs.view(c.num_utter_per_class, c.num_classes_in_batch, -1), 0, 1).reshape(
+                inputs.shape
+            )
+            # ToDo: move it to a unit test
+            # labels_converted = torch.transpose(labels.view(c.num_utter_per_class, c.num_classes_in_batch), 0, 1).reshape(labels.shape)
+            # inputs_converted = torch.transpose(inputs.view(c.num_utter_per_class, c.num_classes_in_batch, -1), 0, 1).reshape(inputs.shape)
+            # idx = 0
+            # for j in range(0, c.num_classes_in_batch, 1):
+            #     for i in range(j, len(labels), c.num_classes_in_batch):
+            #         if not torch.all(labels[i].eq(labels_converted[idx])) or not torch.all(inputs[i].eq(inputs_converted[idx])):
+            #             print("Invalid")
+            #             print(labels)
+            #             exit()
+            #         idx += 1
+            # labels = labels_converted
+            # inputs = inputs_converted
+
+            loader_time = time.time() - end_time
+            global_step += 1
+
+            # setup lr
+            if c.lr_decay:
+                scheduler.step()
+            optimizer.zero_grad()
+
+            # dispatch data to GPU
+            if use_cuda:
+                inputs = inputs.cuda(non_blocking=True)
+                labels = labels.cuda(non_blocking=True)
+
+            # forward pass model
+            outputs = model(inputs)
+
+            # loss computation
+            loss = criterion(
+                outputs.view(c.num_classes_in_batch, outputs.shape[0] // c.num_classes_in_batch, -1), labels
+            )
+            loss.backward()
+            grad_norm, _ = check_update(model, c.grad_clip)
+            optimizer.step()
+
+            step_time = time.time() - start_time
+            epoch_time += step_time
+
+            # acumulate the total epoch loss
+            tot_loss += loss.item()
+
+            # Averaged Loader Time
+            num_loader_workers = c.num_loader_workers if c.num_loader_workers > 0 else 1
+            avg_loader_time = (
+                1 / num_loader_workers * loader_time + (num_loader_workers - 1) / num_loader_workers * avg_loader_time
+                if avg_loader_time != 0
+                else loader_time
+            )
+            current_lr = optimizer.param_groups[0]["lr"]
+
+            if global_step % c.steps_plot_stats == 0:
+                # Plot Training Epoch Stats
+                train_stats = {
+                    "loss": loss.item(),
+                    "lr": current_lr,
+                    "grad_norm": grad_norm,
+                    "step_time": step_time,
+                    "avg_loader_time": avg_loader_time,
+                }
+                dashboard_logger.train_epoch_stats(global_step, train_stats)
+                figures = {
+                    "UMAP Plot": plot_embeddings(outputs.detach().cpu().numpy(), c.num_classes_in_batch),
+                }
+                dashboard_logger.train_figures(global_step, figures)
+
+            if global_step % c.print_step == 0:
+                print(
+                    "   | > Step:{}  Loss:{:.5f}  GradNorm:{:.5f}  "
+                    "StepTime:{:.2f}  LoaderTime:{:.2f}  AvGLoaderTime:{:.2f}  LR:{:.6f}".format(
+                        global_step, loss.item(), grad_norm, step_time, loader_time, avg_loader_time, current_lr
+                    ),
+                    flush=True,
+                )
+
+            if global_step % c.save_step == 0:
+                # save model
+                save_checkpoint(model, optimizer, criterion, loss.item(), OUT_PATH, global_step, epoch)
+
+            end_time = time.time()
+
+        print("")
+        print(
+            ">>> Epoch:{}  AvgLoss: {:.5f} GradNorm:{:.5f}  "
+            "EpochTime:{:.2f} AvGLoaderTime:{:.2f} ".format(
+                epoch, tot_loss / len(data_loader), grad_norm, epoch_time, avg_loader_time
+            ),
+            flush=True,
+        )
+        # evaluation
+        if c.run_eval:
+            model.eval()
+            eval_loss = evaluation(model, criterion, eval_data_loader, global_step)
+            print("\n\n")
+            print("--> EVAL PERFORMANCE")
+            print(
+                "   | > Epoch:{}  AvgLoss: {:.5f} ".format(epoch, eval_loss),
+                flush=True,
+            )
+            # save the best checkpoint
+            best_loss = save_best_model(model, optimizer, criterion, eval_loss, best_loss, OUT_PATH, global_step, epoch)
+            model.train()
+
+    return best_loss, global_step
+
+
+def main(args):  # pylint: disable=redefined-outer-name
+    # pylint: disable=global-variable-undefined
+    global meta_data_train
+    global meta_data_eval
+    global train_classes
+
+    ap = AudioProcessor(**c.audio)
+    model = setup_encoder_model(c)
+
+    optimizer = get_optimizer(c.optimizer, c.optimizer_params, c.lr, model)
+
+    # pylint: disable=redefined-outer-name
+    meta_data_train, meta_data_eval = load_tts_samples(c.datasets, eval_split=True)
+
+    train_data_loader, train_classes, map_classid_to_classname = setup_loader(ap, is_val=False, verbose=True)
+    if c.run_eval:
+        eval_data_loader, _, _ = setup_loader(ap, is_val=True, verbose=True)
+    else:
+        eval_data_loader = None
+
+    num_classes = len(train_classes)
+    criterion = model.get_criterion(c, num_classes)
+
+    if c.loss == "softmaxproto" and c.model != "speaker_encoder":
+        c.map_classid_to_classname = map_classid_to_classname
+        copy_model_files(c, OUT_PATH)
+
+    if args.restore_path:
+        criterion, args.restore_step = model.load_checkpoint(
+            c, args.restore_path, eval=False, use_cuda=use_cuda, criterion=criterion
+        )
+        print(" > Model restored from step %d" % args.restore_step, flush=True)
+    else:
+        args.restore_step = 0
+
+    if c.lr_decay:
+        scheduler = NoamLR(optimizer, warmup_steps=c.warmup_steps, last_epoch=args.restore_step - 1)
+    else:
+        scheduler = None
+
+    num_params = count_parameters(model)
+    print("\n > Model has {} parameters".format(num_params), flush=True)
+
+    if use_cuda:
+        model = model.cuda()
+        criterion.cuda()
+
+    global_step = args.restore_step
+    _, global_step = train(model, optimizer, scheduler, criterion, train_data_loader, eval_data_loader, global_step)
+
+
+if __name__ == "__main__":
+    args, c, OUT_PATH, AUDIO_PATH, c_logger, dashboard_logger = init_training()
+
+    try:
+        main(args)
+    except KeyboardInterrupt:
+        remove_experiment_folder(OUT_PATH)
+        try:
+            sys.exit(0)
+        except SystemExit:
+            os._exit(0)  # pylint: disable=protected-access
+    except Exception:  # pylint: disable=broad-except
+        remove_experiment_folder(OUT_PATH)
+        traceback.print_exc()
+        sys.exit(1)

TTS/bin/train_tts.py

@@ -0,0 +1,75 @@
+import os
+import sys
+
+sys.path.append('.')
+from dataclasses import dataclass, field
+
+from trainer import Trainer, TrainerArgs
+
+from TTS.config import load_config, register_config
+from TTS.tts.datasets import load_tts_samples
+from TTS.tts.models import setup_model
+
+
+@dataclass
+class TrainTTSArgs(TrainerArgs):
+    config_path: str = field(default=None, metadata={"help": "Path to the config file."})
+
+
+def main():
+    os.environ["CUDA_VISIBLE_DEVICES"]="0"
+    """Run `tts` model training directly by a `config.json` file."""
+    # init trainer args
+    train_args = TrainTTSArgs()
+    parser = train_args.init_argparse(arg_prefix="")
+
+    # override trainer args from comman-line args
+    args, config_overrides = parser.parse_known_args()
+    train_args.parse_args(args)
+
+    # load config.json and register
+    if args.config_path or args.continue_path:
+        if args.config_path:
+            # init from a file
+            config = load_config(args.config_path)
+            if len(config_overrides) > 0:
+                config.parse_known_args(config_overrides, relaxed_parser=True)
+        elif args.continue_path:
+            # continue from a prev experiment
+            config = load_config(os.path.join(args.continue_path, "config.json"))
+            if len(config_overrides) > 0:
+                config.parse_known_args(config_overrides, relaxed_parser=True)
+        else:
+            # init from console args
+            from TTS.config.shared_configs import BaseTrainingConfig  # pylint: disable=import-outside-toplevel
+
+            config_base = BaseTrainingConfig()
+            config_base.parse_known_args(config_overrides)
+            config = register_config(config_base.model)()
+
+    # load training samples
+    train_samples, eval_samples = load_tts_samples(
+        config.datasets,
+        eval_split=True,
+        eval_split_max_size=config.eval_split_max_size,
+        eval_split_size=config.eval_split_size,
+    )
+
+    # init the model from config
+    model = setup_model(config, train_samples + eval_samples)
+
+    # init the trainer and 🚀
+    trainer = Trainer(
+        train_args,
+        model.config,
+        config.output_path,
+        model=model,
+        train_samples=train_samples,
+        eval_samples=eval_samples,
+        parse_command_line_args=False,
+    )
+    trainer.fit()
+
+
+if __name__ == "__main__":
+    main()

TTS/bin/train_vocoder.py

@@ -0,0 +1,81 @@
+import os
+from dataclasses import dataclass, field
+import sys
+
+sys.path.append('.')
+from trainer import Trainer, TrainerArgs
+import torch
+from TTS.config import load_config, register_config
+from TTS.utils.audio import AudioProcessor
+from TTS.vocoder.datasets.preprocess import load_wav_data, load_wav_feat_data
+from TTS.vocoder.models import setup_model
+
+
+@dataclass
+class TrainVocoderArgs(TrainerArgs):
+    config_path: str = field(default=None, metadata={"help": "Path to the config file."})
+
+
+def main():
+  
+    os.environ["CUDA_VISIBLE_DEVICES"]="0"
+    """Run `tts` model training directly by a `config.json` file."""
+    # init trainer args
+    train_args = TrainVocoderArgs()
+    parser = train_args.init_argparse(arg_prefix="")
+
+    # override trainer args from comman-line args
+    args, config_overrides = parser.parse_known_args()
+    train_args.parse_args(args)
+
+    # load config.json and register
+    if args.config_path or args.continue_path:
+        if args.config_path:
+            # init from a file
+            config = load_config(args.config_path)
+            if len(config_overrides) > 0:
+                config.parse_known_args(config_overrides, relaxed_parser=True)
+        elif args.continue_path:
+            # continue from a prev experiment
+            config = load_config(os.path.join(args.continue_path, "config.json"))
+            if len(config_overrides) > 0:
+                config.parse_known_args(config_overrides, relaxed_parser=True)
+        else:
+            # init from console args
+            from TTS.config.shared_configs import BaseTrainingConfig  # pylint: disable=import-outside-toplevel
+
+            config_base = BaseTrainingConfig()
+            config_base.parse_known_args(config_overrides)
+            config = register_config(config_base.model)()
+
+    # load training samples
+    if "feature_path" in config and config.feature_path:
+        # load pre-computed features
+        print(f" > Loading features from: {config.feature_path}")
+        eval_samples, train_samples = load_wav_feat_data(config.data_path, config.feature_path, config.eval_split_size)
+    else:
+        # load data raw wav files
+        eval_samples, train_samples = load_wav_data(config.data_path, config.eval_split_size)
+
+    # setup audio processor
+    ap = AudioProcessor(**config.audio)
+
+    # init the model from config
+    model = setup_model(config)
+
+    # init the trainer and 🚀
+    trainer = Trainer(
+        train_args,
+        config,
+        config.output_path,
+        model=model,
+        train_samples=train_samples,
+        eval_samples=eval_samples,
+        training_assets={"audio_processor": ap},
+        parse_command_line_args=False,
+    )
+    trainer.fit()
+
+
+if __name__ == "__main__":
+    main()

TTS/bin/tune_wavegrad.py

@@ -0,0 +1,103 @@
+"""Search a good noise schedule for WaveGrad for a given number of inference iterations"""
+import argparse
+from itertools import product as cartesian_product
+
+import numpy as np
+import torch
+from torch.utils.data import DataLoader
+from tqdm import tqdm
+
+from TTS.config import load_config
+from TTS.utils.audio import AudioProcessor
+from TTS.vocoder.datasets.preprocess import load_wav_data
+from TTS.vocoder.datasets.wavegrad_dataset import WaveGradDataset
+from TTS.vocoder.models import setup_model
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model_path", type=str, help="Path to model checkpoint.")
+    parser.add_argument("--config_path", type=str, help="Path to model config file.")
+    parser.add_argument("--data_path", type=str, help="Path to data directory.")
+    parser.add_argument("--output_path", type=str, help="path for output file including file name and extension.")
+    parser.add_argument(
+        "--num_iter",
+        type=int,
+        help="Number of model inference iterations that you like to optimize noise schedule for.",
+    )
+    parser.add_argument("--use_cuda", action="store_true", help="enable CUDA.")
+    parser.add_argument("--num_samples", type=int, default=1, help="Number of datasamples used for inference.")
+    parser.add_argument(
+        "--search_depth",
+        type=int,
+        default=3,
+        help="Search granularity. Increasing this increases the run-time exponentially.",
+    )
+
+    # load config
+    args = parser.parse_args()
+    config = load_config(args.config_path)
+
+    # setup audio processor
+    ap = AudioProcessor(**config.audio)
+
+    # load dataset
+    _, train_data = load_wav_data(args.data_path, 0)
+    train_data = train_data[: args.num_samples]
+    dataset = WaveGradDataset(
+        ap=ap,
+        items=train_data,
+        seq_len=-1,
+        hop_len=ap.hop_length,
+        pad_short=config.pad_short,
+        conv_pad=config.conv_pad,
+        is_training=True,
+        return_segments=False,
+        use_noise_augment=False,
+        use_cache=False,
+        verbose=True,
+    )
+    loader = DataLoader(
+        dataset,
+        batch_size=1,
+        shuffle=False,
+        collate_fn=dataset.collate_full_clips,
+        drop_last=False,
+        num_workers=config.num_loader_workers,
+        pin_memory=False,
+    )
+
+    # setup the model
+    model = setup_model(config)
+    if args.use_cuda:
+        model.cuda()
+
+    # setup optimization parameters
+    base_values = sorted(10 * np.random.uniform(size=args.search_depth))
+    print(f" > base values: {base_values}")
+    exponents = 10 ** np.linspace(-6, -1, num=args.num_iter)
+    best_error = float("inf")
+    best_schedule = None  # pylint: disable=C0103
+    total_search_iter = len(base_values) ** args.num_iter
+    for base in tqdm(cartesian_product(base_values, repeat=args.num_iter), total=total_search_iter):
+        beta = exponents * base
+        model.compute_noise_level(beta)
+        for data in loader:
+            mel, audio = data
+            y_hat = model.inference(mel.cuda() if args.use_cuda else mel)
+
+            if args.use_cuda:
+                y_hat = y_hat.cpu()
+            y_hat = y_hat.numpy()
+
+            mel_hat = []
+            for i in range(y_hat.shape[0]):
+                m = ap.melspectrogram(y_hat[i, 0])[:, :-1]
+                mel_hat.append(torch.from_numpy(m))
+
+            mel_hat = torch.stack(mel_hat)
+            mse = torch.sum((mel - mel_hat) ** 2).mean()
+            if mse.item() < best_error:
+                best_error = mse.item()
+                best_schedule = {"beta": beta}
+                print(f" > Found a better schedule. - MSE: {mse.item()}")
+                np.save(args.output_path, best_schedule)

TTS/config/.ipynb_checkpoints/__init__-checkpoint.py

@@ -0,0 +1,132 @@
+import json
+import os
+import re
+from typing import Dict
+
+import fsspec
+import yaml
+from coqpit import Coqpit
+
+from TTS.config.shared_configs import *
+from TTS.utils.generic_utils import find_module
+
+
+def read_json_with_comments(json_path):
+    """for backward compat."""
+    # fallback to json
+    with fsspec.open(json_path, "r", encoding="utf-8") as f:
+        input_str = f.read()
+    # handle comments
+    input_str = re.sub(r"\\\n", "", input_str)
+    input_str = re.sub(r"//.*\n", "\n", input_str)
+    data = json.loads(input_str)
+    return data
+
+
+def register_config(model_name: str) -> Coqpit:
+    """Find the right config for the given model name.
+
+    Args:
+        model_name (str): Model name.
+
+    Raises:
+        ModuleNotFoundError: No matching config for the model name.
+
+    Returns:
+        Coqpit: config class.
+    """
+    config_class = None
+    config_name = model_name + "_config"
+    paths = ["TTS.tts.configs", "TTS.vocoder.configs", "TTS.encoder.configs"]
+    for path in paths:
+        try:
+            config_class = find_module(path, config_name)
+        except ModuleNotFoundError:
+            pass
+    if config_class is None:
+        raise ModuleNotFoundError(f" [!] Config for {model_name} cannot be found.")
+    return config_class
+
+
+def _process_model_name(config_dict: Dict) -> str:
+    """Format the model name as expected. It is a band-aid for the old `vocoder` model names.
+
+    Args:
+        config_dict (Dict): A dictionary including the config fields.
+
+    Returns:
+        str: Formatted modelname.
+    """
+    model_name = config_dict["model"] if "model" in config_dict else config_dict["generator_model"]
+    model_name = model_name.replace("_generator", "").replace("_discriminator", "")
+    return model_name
+
+
+def load_config(config_path: str) -> Coqpit:
+    """Import `json` or `yaml` files as TTS configs. First, load the input file as a `dict` and check the model name
+    to find the corresponding Config class. Then initialize the Config.
+
+    Args:
+        config_path (str): path to the config file.
+
+    Raises:
+        TypeError: given config file has an unknown type.
+
+    Returns:
+        Coqpit: TTS config object.
+    """
+    config_dict = {}
+    ext = os.path.splitext(config_path)[1]
+    if ext in (".yml", ".yaml"):
+        with fsspec.open(config_path, "r", encoding="utf-8") as f:
+            data = yaml.safe_load(f)
+    elif ext == ".json":
+        try:
+            with fsspec.open(config_path, "r", encoding="utf-8") as f:
+                data = json.load(f)
+        except json.decoder.JSONDecodeError:
+            # backwards compat.
+            data = read_json_with_comments(config_path)
+    else:
+        raise TypeError(f" [!] Unknown config file type {ext}")
+    config_dict.update(data)
+    model_name = _process_model_name(config_dict)
+    config_class = register_config(model_name.lower())
+    config = config_class()
+    config.from_dict(config_dict)
+    return config
+
+
+def check_config_and_model_args(config, arg_name, value):
+    """Check the give argument in `config.model_args` if exist or in `config` for
+    the given value.
+
+    Return False if the argument does not exist in `config.model_args` or `config`.
+    This is to patch up the compatibility between models with and without `model_args`.
+
+    TODO: Remove this in the future with a unified approach.
+    """
+    if hasattr(config, "model_args"):
+        if arg_name in config.model_args:
+            return config.model_args[arg_name] == value
+    if hasattr(config, arg_name):
+        return config[arg_name] == value
+    return False
+
+
+def get_from_config_or_model_args(config, arg_name):
+    """Get the given argument from `config.model_args` if exist or in `config`."""
+    if hasattr(config, "model_args"):
+        if arg_name in config.model_args:
+            return config.model_args[arg_name]
+    return config[arg_name]
+
+
+def get_from_config_or_model_args_with_default(config, arg_name, def_val):
+    """Get the given argument from `config.model_args` if exist or in `config`."""
+    if hasattr(config, "model_args"):
+        if arg_name in config.model_args:
+            return config.model_args[arg_name]
+    if hasattr(config, arg_name):
+        return config[arg_name]
+    return def_val

TTS/config/.ipynb_checkpoints/config-checkpoint.json

@@ -0,0 +1,20 @@
+{
+   "model": "glow_tts",
+   "batch_size": 32,
+   "eval_batch_size": 16,
+   "num_loader_workers": 4,
+   "num_eval_loader_workers": 4,
+   "run_eval": true,
+   "test_delay_epochs": -1,
+   "epochs": 1000,
+   "text_cleaner": "english_cleaners",
+   "use_phonemes": false,
+   "phoneme_language": "en-us",
+   "phoneme_cache_path": "phoneme_cache",
+   "print_step": 25,
+   "print_eval": true,
+   "mixed_precision": false,
+   "output_path": "recipes/ljspeech/glow_tts/",
+   "test_sentences": ["Test this sentence.", "This test sentence.", "Sentence this test."],
+   "datasets":[{"formatter": "infore", "meta_file_train":"scripts.csv", "path":"/Users/saltlux/Code/SpeechSynthesis/Dataset/25hours/"}]
+}

TTS/config/.ipynb_checkpoints/shared_configs-checkpoint.py

@@ -0,0 +1,264 @@
+from dataclasses import asdict, dataclass
+from typing import List
+
+from coqpit import Coqpit, check_argument
+from trainer import TrainerConfig
+
+
+@dataclass
+class BaseAudioConfig(Coqpit):
+    """Base config to definge audio processing parameters. It is used to initialize
+    ```TTS.utils.audio.AudioProcessor.```
+
+    Args:
+        fft_size (int):
+            Number of STFT frequency levels aka.size of the linear spectogram frame. Defaults to 1024.
+
+        win_length (int):
+            Each frame of audio is windowed by window of length ```win_length``` and then padded with zeros to match
+            ```fft_size```. Defaults to 1024.
+
+        hop_length (int):
+            Number of audio samples between adjacent STFT columns. Defaults to 1024.
+
+        frame_shift_ms (int):
+            Set ```hop_length``` based on milliseconds and sampling rate.
+
+        frame_length_ms (int):
+            Set ```win_length``` based on milliseconds and sampling rate.
+
+        stft_pad_mode (str):
+            Padding method used in STFT. 'reflect' or 'center'. Defaults to 'reflect'.
+
+        sample_rate (int):
+            Audio sampling rate. Defaults to 22050.
+
+        resample (bool):
+            Enable / Disable resampling audio to ```sample_rate```. Defaults to ```False```.
+
+        preemphasis (float):
+            Preemphasis coefficient. Defaults to 0.0.
+
+        ref_level_db (int): 20
+            Reference Db level to rebase the audio signal and ignore the level below. 20Db is assumed the sound of air.
+            Defaults to 20.
+
+        do_sound_norm (bool):
+            Enable / Disable sound normalization to reconcile the volume differences among samples. Defaults to False.
+
+        log_func (str):
+            Numpy log function used for amplitude to DB conversion. Defaults to 'np.log10'.
+
+        do_trim_silence (bool):
+            Enable / Disable trimming silences at the beginning and the end of the audio clip. Defaults to ```True```.
+
+        do_amp_to_db_linear (bool, optional):
+            enable/disable amplitude to dB conversion of linear spectrograms. Defaults to True.
+
+        do_amp_to_db_mel (bool, optional):
+            enable/disable amplitude to dB conversion of mel spectrograms. Defaults to True.
+
+        pitch_fmax (float, optional):
+            Maximum frequency of the F0 frames. Defaults to ```640```.
+
+        pitch_fmin (float, optional):
+            Minimum frequency of the F0 frames. Defaults to ```1```.
+
+        trim_db (int):
+            Silence threshold used for silence trimming. Defaults to 45.
+
+        do_rms_norm (bool, optional):
+            enable/disable RMS volume normalization when loading an audio file. Defaults to False.
+
+        db_level (int, optional):
+            dB level used for rms normalization. The range is -99 to 0. Defaults to None.
+
+        power (float):
+            Exponent used for expanding spectrogra levels before running Griffin Lim. It helps to reduce the
+            artifacts in the synthesized voice. Defaults to 1.5.
+
+        griffin_lim_iters (int):
+            Number of Griffing Lim iterations. Defaults to 60.
+
+        num_mels (int):
+            Number of mel-basis frames that defines the frame lengths of each mel-spectrogram frame. Defaults to 80.
+
+        mel_fmin (float): Min frequency level used for the mel-basis filters. ~50 for male and ~95 for female voices.
+            It needs to be adjusted for a dataset. Defaults to 0.
+
+        mel_fmax (float):
+            Max frequency level used for the mel-basis filters. It needs to be adjusted for a dataset.
+
+        spec_gain (int):
+            Gain applied when converting amplitude to DB. Defaults to 20.
+
+        signal_norm (bool):
+            enable/disable signal normalization. Defaults to True.
+
+        min_level_db (int):
+            minimum db threshold for the computed melspectrograms. Defaults to -100.
+
+        symmetric_norm (bool):
+            enable/disable symmetric normalization. If set True normalization is performed in the range [-k, k] else
+            [0, k], Defaults to True.
+
+        max_norm (float):
+            ```k``` defining the normalization range. Defaults to 4.0.
+
+        clip_norm (bool):
+            enable/disable clipping the our of range values in the normalized audio signal. Defaults to True.
+
+        stats_path (str):
+            Path to the computed stats file. Defaults to None.
+    """
+
+    # stft parameters
+    fft_size: int = 1024
+    win_length: int = 1024
+    hop_length: int = 256
+    frame_shift_ms: int = None
+    frame_length_ms: int = None
+    stft_pad_mode: str = "reflect"
+    # audio processing parameters
+    sample_rate: int = 22050
+    resample: bool = False
+    preemphasis: float = 0.0
+    ref_level_db: int = 20
+    do_sound_norm: bool = False
+    log_func: str = "np.log10"
+    # silence trimming
+    do_trim_silence: bool = True
+    trim_db: int = 45
+    # rms volume normalization
+    do_rms_norm: bool = False
+    db_level: float = None
+    # griffin-lim params
+    power: float = 1.5
+    griffin_lim_iters: int = 60
+    # mel-spec params
+    num_mels: int = 80
+    mel_fmin: float = 0.0
+    mel_fmax: float = None
+    spec_gain: int = 20
+    do_amp_to_db_linear: bool = True
+    do_amp_to_db_mel: bool = True
+    # f0 params
+    pitch_fmax: float = 640.0
+    pitch_fmin: float = 1.0
+    # normalization params
+    signal_norm: bool = True
+    min_level_db: int = -100
+    symmetric_norm: bool = True
+    max_norm: float = 4.0
+    clip_norm: bool = True
+    stats_path: str = None
+
+    def check_values(
+        self,
+    ):
+        """Check config fields"""
+        c = asdict(self)
+        check_argument("num_mels", c, restricted=True, min_val=10, max_val=2056)
+        check_argument("fft_size", c, restricted=True, min_val=128, max_val=4058)
+        check_argument("sample_rate", c, restricted=True, min_val=512, max_val=100000)
+        check_argument(
+            "frame_length_ms",
+            c,
+            restricted=True,
+            min_val=10,
+            max_val=1000,
+            alternative="win_length",
+        )
+        check_argument("frame_shift_ms", c, restricted=True, min_val=1, max_val=1000, alternative="hop_length")
+        check_argument("preemphasis", c, restricted=True, min_val=0, max_val=1)
+        check_argument("min_level_db", c, restricted=True, min_val=-1000, max_val=10)
+        check_argument("ref_level_db", c, restricted=True, min_val=0, max_val=1000)
+        check_argument("power", c, restricted=True, min_val=1, max_val=5)
+        check_argument("griffin_lim_iters", c, restricted=True, min_val=10, max_val=1000)
+
+        # normalization parameters
+        check_argument("signal_norm", c, restricted=True)
+        check_argument("symmetric_norm", c, restricted=True)
+        check_argument("max_norm", c, restricted=True, min_val=0.1, max_val=1000)
+        check_argument("clip_norm", c, restricted=True)
+        check_argument("mel_fmin", c, restricted=True, min_val=0.0, max_val=1000)
+        check_argument("mel_fmax", c, restricted=True, min_val=500.0, allow_none=True)
+        check_argument("spec_gain", c, restricted=True, min_val=1, max_val=100)
+        check_argument("do_trim_silence", c, restricted=True)
+        check_argument("trim_db", c, restricted=True)
+
+
+@dataclass
+class BaseDatasetConfig(Coqpit):
+    """Base config for TTS datasets.
+
+    Args:
+        formatter (str):
+            Formatter name that defines used formatter in ```TTS.tts.datasets.formatter```. Defaults to `""`.
+
+        dataset_name (str):
+            Unique name for the dataset. Defaults to `""`.
+
+        path (str):
+            Root path to the dataset files. Defaults to `""`.
+
+        meta_file_train (str):
+            Name of the dataset meta file. Or a list of speakers to be ignored at training for multi-speaker datasets.
+            Defaults to `""`.
+
+        ignored_speakers (List):
+            List of speakers IDs that are not used at the training. Default None.
+
+        language (str):
+            Language code of the dataset. If defined, it overrides `phoneme_language`. Defaults to `""`.
+
+        meta_file_val (str):
+            Name of the dataset meta file that defines the instances used at validation.
+
+        meta_file_attn_mask (str):
+            Path to the file that lists the attention mask files used with models that require attention masks to
+            train the duration predictor.
+    """
+
+    formatter: str = ""
+    dataset_name: str = ""
+    path: str = ""
+    meta_file_train: str = ""
+    ignored_speakers: List[str] = None
+    language: str = ""
+    meta_file_val: str = ""
+    meta_file_attn_mask: str = ""
+
+    def check_values(
+        self,
+    ):
+        """Check config fields"""
+        c = asdict(self)
+        check_argument("formatter", c, restricted=True)
+        check_argument("path", c, restricted=True)
+        check_argument("meta_file_train", c, restricted=True)
+        check_argument("meta_file_val", c, restricted=False)
+        check_argument("meta_file_attn_mask", c, restricted=False)
+
+
+@dataclass
+class BaseTrainingConfig(TrainerConfig):
+    """Base config to define the basic 🐸TTS training parameters that are shared
+    among all the models. It is based on ```Trainer.TrainingConfig```.
+
+    Args:
+        model (str):
+            Name of the model that is used in the training.
+
+        num_loader_workers (int):
+            Number of workers for training time dataloader.
+
+        num_eval_loader_workers (int):
+            Number of workers for evaluation time dataloader.
+    """
+
+    model: str = None
+    # dataloading
+    num_loader_workers: int = 0
+    num_eval_loader_workers: int = 0
+    use_noise_augment: bool = False

TTS/config/__init__.py

@@ -0,0 +1,132 @@
+import json
+import os
+import re
+from typing import Dict
+
+import fsspec
+import yaml
+from coqpit import Coqpit
+
+from TTS.config.shared_configs import *
+from TTS.utils.generic_utils import find_module
+
+
+def read_json_with_comments(json_path):
+    """for backward compat."""
+    # fallback to json
+    with fsspec.open(json_path, "r", encoding="utf-8") as f:
+        input_str = f.read()
+    # handle comments
+    input_str = re.sub(r"\\\n", "", input_str)
+    input_str = re.sub(r"//.*\n", "\n", input_str)
+    data = json.loads(input_str)
+    return data
+
+
+def register_config(model_name: str) -> Coqpit:
+    """Find the right config for the given model name.
+
+    Args:
+        model_name (str): Model name.
+
+    Raises:
+        ModuleNotFoundError: No matching config for the model name.
+
+    Returns:
+        Coqpit: config class.
+    """
+    config_class = None
+    config_name = model_name + "_config"
+    paths = ["TTS.tts.configs", "TTS.vocoder.configs", "TTS.encoder.configs"]
+    for path in paths:
+        try:
+            config_class = find_module(path, config_name)
+        except ModuleNotFoundError:
+            pass
+    if config_class is None:
+        raise ModuleNotFoundError(f" [!] Config for {model_name} cannot be found.")
+    return config_class
+
+
+def _process_model_name(config_dict: Dict) -> str:
+    """Format the model name as expected. It is a band-aid for the old `vocoder` model names.
+
+    Args:
+        config_dict (Dict): A dictionary including the config fields.
+
+    Returns:
+        str: Formatted modelname.
+    """
+    model_name = config_dict["model"] if "model" in config_dict else config_dict["generator_model"]
+    model_name = model_name.replace("_generator", "").replace("_discriminator", "")
+    return model_name
+
+
+def load_config(config_path: str) -> Coqpit:
+    """Import `json` or `yaml` files as TTS configs. First, load the input file as a `dict` and check the model name
+    to find the corresponding Config class. Then initialize the Config.
+
+    Args:
+        config_path (str): path to the config file.
+
+    Raises:
+        TypeError: given config file has an unknown type.
+
+    Returns:
+        Coqpit: TTS config object.
+    """
+    config_dict = {}
+    ext = os.path.splitext(config_path)[1]
+    if ext in (".yml", ".yaml"):
+        with fsspec.open(config_path, "r", encoding="utf-8") as f:
+            data = yaml.safe_load(f)
+    elif ext == ".json":
+        try:
+            with fsspec.open(config_path, "r", encoding="utf-8") as f:
+                data = json.load(f)
+        except json.decoder.JSONDecodeError:
+            # backwards compat.
+            data = read_json_with_comments(config_path)
+    else:
+        raise TypeError(f" [!] Unknown config file type {ext}")
+    config_dict.update(data)
+    model_name = _process_model_name(config_dict)
+    config_class = register_config(model_name.lower())
+    config = config_class()
+    config.from_dict(config_dict)
+    return config
+
+
+def check_config_and_model_args(config, arg_name, value):
+    """Check the give argument in `config.model_args` if exist or in `config` for
+    the given value.
+
+    Return False if the argument does not exist in `config.model_args` or `config`.
+    This is to patch up the compatibility between models with and without `model_args`.
+
+    TODO: Remove this in the future with a unified approach.
+    """
+    if hasattr(config, "model_args"):
+        if arg_name in config.model_args:
+            return config.model_args[arg_name] == value
+    if hasattr(config, arg_name):
+        return config[arg_name] == value
+    return False
+
+
+def get_from_config_or_model_args(config, arg_name):
+    """Get the given argument from `config.model_args` if exist or in `config`."""
+    if hasattr(config, "model_args"):
+        if arg_name in config.model_args:
+            return config.model_args[arg_name]
+    return config[arg_name]
+
+
+def get_from_config_or_model_args_with_default(config, arg_name, def_val):
+    """Get the given argument from `config.model_args` if exist or in `config`."""
+    if hasattr(config, "model_args"):
+        if arg_name in config.model_args:
+            return config.model_args[arg_name]
+    if hasattr(config, arg_name):
+        return config[arg_name]
+    return def_val

TTS/config/config.json

@@ -0,0 +1,25 @@
+{
+   "model": "glow_tts",
+   "batch_size": 32,
+   "eval_batch_size": 16,
+   "num_loader_workers": 4,
+   "num_eval_loader_workers": 4,
+   "run_eval": true,
+   "test_delay_epochs": -1,
+   "epochs": 1000,
+   "text_cleaner": "english_cleaners",
+   "use_phonemes": false,
+   "phoneme_language": "en-us",
+   "phoneme_cache_path": "phoneme_cache",
+   "print_step": 25,
+   "print_eval": true,
+   "mixed_precision": false,
+   "output_path": "recipes/ljspeech/glow_tts/",
+   "test_sentences": ["Test this sentence.", "This test sentence.", "Sentence this test."],
+   "datasets":[{"formatter": "infore", "meta_file_train":"scripts.csv", "path":"/Users/saltlux/Code/SpeechSynthesis/Dataset/25hours/"}],
+   "characters":{
+        "characters": "abcdeghiklmnopqrstuvxyàáâãèéêìíòóôõùúýăđĩũơưạảấầẩẫậắằẳẵặẹẻẽếềểễệỉịọỏốồổỗộớờởỡợụủứừửữựỳỵỷỹ",
+        "phonemes": null,
+        "unique": true
+    }
+}

TTS/config/shared_configs.py

@@ -0,0 +1,264 @@
+from dataclasses import asdict, dataclass
+from typing import List
+
+from coqpit import Coqpit, check_argument
+from trainer import TrainerConfig
+
+
+@dataclass
+class BaseAudioConfig(Coqpit):
+    """Base config to definge audio processing parameters. It is used to initialize
+    ```TTS.utils.audio.AudioProcessor.```
+
+    Args:
+        fft_size (int):
+            Number of STFT frequency levels aka.size of the linear spectogram frame. Defaults to 1024.
+
+        win_length (int):
+            Each frame of audio is windowed by window of length ```win_length``` and then padded with zeros to match
+            ```fft_size```. Defaults to 1024.
+
+        hop_length (int):
+            Number of audio samples between adjacent STFT columns. Defaults to 1024.
+
+        frame_shift_ms (int):
+            Set ```hop_length``` based on milliseconds and sampling rate.
+
+        frame_length_ms (int):
+            Set ```win_length``` based on milliseconds and sampling rate.
+
+        stft_pad_mode (str):
+            Padding method used in STFT. 'reflect' or 'center'. Defaults to 'reflect'.
+
+        sample_rate (int):
+            Audio sampling rate. Defaults to 22050.
+
+        resample (bool):
+            Enable / Disable resampling audio to ```sample_rate```. Defaults to ```False```.
+
+        preemphasis (float):
+            Preemphasis coefficient. Defaults to 0.0.
+
+        ref_level_db (int): 20
+            Reference Db level to rebase the audio signal and ignore the level below. 20Db is assumed the sound of air.
+            Defaults to 20.
+
+        do_sound_norm (bool):
+            Enable / Disable sound normalization to reconcile the volume differences among samples. Defaults to False.
+
+        log_func (str):
+            Numpy log function used for amplitude to DB conversion. Defaults to 'np.log10'.
+
+        do_trim_silence (bool):
+            Enable / Disable trimming silences at the beginning and the end of the audio clip. Defaults to ```True```.
+
+        do_amp_to_db_linear (bool, optional):
+            enable/disable amplitude to dB conversion of linear spectrograms. Defaults to True.
+
+        do_amp_to_db_mel (bool, optional):
+            enable/disable amplitude to dB conversion of mel spectrograms. Defaults to True.
+
+        pitch_fmax (float, optional):
+            Maximum frequency of the F0 frames. Defaults to ```640```.
+
+        pitch_fmin (float, optional):
+            Minimum frequency of the F0 frames. Defaults to ```1```.
+
+        trim_db (int):
+            Silence threshold used for silence trimming. Defaults to 45.
+
+        do_rms_norm (bool, optional):
+            enable/disable RMS volume normalization when loading an audio file. Defaults to False.
+
+        db_level (int, optional):
+            dB level used for rms normalization. The range is -99 to 0. Defaults to None.
+
+        power (float):
+            Exponent used for expanding spectrogra levels before running Griffin Lim. It helps to reduce the
+            artifacts in the synthesized voice. Defaults to 1.5.
+
+        griffin_lim_iters (int):
+            Number of Griffing Lim iterations. Defaults to 60.
+
+        num_mels (int):
+            Number of mel-basis frames that defines the frame lengths of each mel-spectrogram frame. Defaults to 80.
+
+        mel_fmin (float): Min frequency level used for the mel-basis filters. ~50 for male and ~95 for female voices.
+            It needs to be adjusted for a dataset. Defaults to 0.
+
+        mel_fmax (float):
+            Max frequency level used for the mel-basis filters. It needs to be adjusted for a dataset.
+
+        spec_gain (int):
+            Gain applied when converting amplitude to DB. Defaults to 20.
+
+        signal_norm (bool):
+            enable/disable signal normalization. Defaults to True.
+
+        min_level_db (int):
+            minimum db threshold for the computed melspectrograms. Defaults to -100.
+
+        symmetric_norm (bool):
+            enable/disable symmetric normalization. If set True normalization is performed in the range [-k, k] else
+            [0, k], Defaults to True.
+
+        max_norm (float):
+            ```k``` defining the normalization range. Defaults to 4.0.
+
+        clip_norm (bool):
+            enable/disable clipping the our of range values in the normalized audio signal. Defaults to True.
+
+        stats_path (str):
+            Path to the computed stats file. Defaults to None.
+    """
+
+    # stft parameters
+    fft_size: int = 1024
+    win_length: int = 1024
+    hop_length: int = 256
+    frame_shift_ms: int = None
+    frame_length_ms: int = None
+    stft_pad_mode: str = "reflect"
+    # audio processing parameters
+    sample_rate: int = 22050
+    resample: bool = False
+    preemphasis: float = 0.0
+    ref_level_db: int = 20
+    do_sound_norm: bool = False
+    log_func: str = "np.log10"
+    # silence trimming
+    do_trim_silence: bool = True
+    trim_db: int = 45
+    # rms volume normalization
+    do_rms_norm: bool = False
+    db_level: float = None
+    # griffin-lim params
+    power: float = 1.5
+    griffin_lim_iters: int = 60
+    # mel-spec params
+    num_mels: int = 80
+    mel_fmin: float = 0.0
+    mel_fmax: float = None
+    spec_gain: int = 20
+    do_amp_to_db_linear: bool = True
+    do_amp_to_db_mel: bool = True
+    # f0 params
+    pitch_fmax: float = 640.0
+    pitch_fmin: float = 1.0
+    # normalization params
+    signal_norm: bool = True
+    min_level_db: int = -100
+    symmetric_norm: bool = True
+    max_norm: float = 4.0
+    clip_norm: bool = True
+    stats_path: str = None
+
+    def check_values(
+        self,
+    ):
+        """Check config fields"""
+        c = asdict(self)
+        check_argument("num_mels", c, restricted=True, min_val=10, max_val=2056)
+        check_argument("fft_size", c, restricted=True, min_val=128, max_val=4058)
+        check_argument("sample_rate", c, restricted=True, min_val=512, max_val=100000)
+        check_argument(
+            "frame_length_ms",
+            c,
+            restricted=True,
+            min_val=10,
+            max_val=1000,
+            alternative="win_length",
+        )
+        check_argument("frame_shift_ms", c, restricted=True, min_val=1, max_val=1000, alternative="hop_length")
+        check_argument("preemphasis", c, restricted=True, min_val=0, max_val=1)
+        check_argument("min_level_db", c, restricted=True, min_val=-1000, max_val=10)
+        check_argument("ref_level_db", c, restricted=True, min_val=0, max_val=1000)
+        check_argument("power", c, restricted=True, min_val=1, max_val=5)
+        check_argument("griffin_lim_iters", c, restricted=True, min_val=10, max_val=1000)
+
+        # normalization parameters
+        check_argument("signal_norm", c, restricted=True)
+        check_argument("symmetric_norm", c, restricted=True)
+        check_argument("max_norm", c, restricted=True, min_val=0.1, max_val=1000)
+        check_argument("clip_norm", c, restricted=True)
+        check_argument("mel_fmin", c, restricted=True, min_val=0.0, max_val=1000)
+        check_argument("mel_fmax", c, restricted=True, min_val=500.0, allow_none=True)
+        check_argument("spec_gain", c, restricted=True, min_val=1, max_val=100)
+        check_argument("do_trim_silence", c, restricted=True)
+        check_argument("trim_db", c, restricted=True)
+
+
+@dataclass
+class BaseDatasetConfig(Coqpit):
+    """Base config for TTS datasets.
+
+    Args:
+        formatter (str):
+            Formatter name that defines used formatter in ```TTS.tts.datasets.formatter```. Defaults to `""`.
+
+        dataset_name (str):
+            Unique name for the dataset. Defaults to `""`.
+
+        path (str):
+            Root path to the dataset files. Defaults to `""`.
+
+        meta_file_train (str):
+            Name of the dataset meta file. Or a list of speakers to be ignored at training for multi-speaker datasets.
+            Defaults to `""`.
+
+        ignored_speakers (List):
+            List of speakers IDs that are not used at the training. Default None.
+
+        language (str):
+            Language code of the dataset. If defined, it overrides `phoneme_language`. Defaults to `""`.
+
+        meta_file_val (str):
+            Name of the dataset meta file that defines the instances used at validation.
+
+        meta_file_attn_mask (str):
+            Path to the file that lists the attention mask files used with models that require attention masks to
+            train the duration predictor.
+    """
+
+    formatter: str = ""
+    dataset_name: str = ""
+    path: str = ""
+    meta_file_train: str = ""
+    ignored_speakers: List[str] = None
+    language: str = ""
+    meta_file_val: str = ""
+    meta_file_attn_mask: str = ""
+
+    def check_values(
+        self,
+    ):
+        """Check config fields"""
+        c = asdict(self)
+        check_argument("formatter", c, restricted=True)
+        check_argument("path", c, restricted=True)
+        check_argument("meta_file_train", c, restricted=True)
+        check_argument("meta_file_val", c, restricted=False)
+        check_argument("meta_file_attn_mask", c, restricted=False)
+
+
+@dataclass
+class BaseTrainingConfig(TrainerConfig):
+    """Base config to define the basic 🐸TTS training parameters that are shared
+    among all the models. It is based on ```Trainer.TrainingConfig```.
+
+    Args:
+        model (str):
+            Name of the model that is used in the training.
+
+        num_loader_workers (int):
+            Number of workers for training time dataloader.
+
+        num_eval_loader_workers (int):
+            Number of workers for evaluation time dataloader.
+    """
+
+    model: str = None
+    # dataloading
+    num_loader_workers: int = 0
+    num_eval_loader_workers: int = 0
+    use_noise_augment: bool = False

TTS/encoder/README.md

@@ -0,0 +1,18 @@
+### Speaker Encoder
+
+This is an implementation of https://arxiv.org/abs/1710.10467. This model can be used for voice and speaker embedding.
+
+With the code here you can generate d-vectors for both multi-speaker and single-speaker TTS datasets, then visualise and explore them along with the associated audio files in an interactive chart.
+
+Below is an example showing embedding results of various speakers. You can generate the same plot with the provided notebook as demonstrated in [this video](https://youtu.be/KW3oO7JVa7Q).
+
+![](umap.png)
+
+Download a pretrained model from [Released Models](https://github.com/mozilla/TTS/wiki/Released-Models) page.
+
+To run the code, you need to follow the same flow as in TTS.
+
+- Define 'config.json' for your needs. Note that, audio parameters should match your TTS model.
+- Example training call ```python speaker_encoder/train.py --config_path speaker_encoder/config.json --data_path ~/Data/Libri-TTS/train-clean-360```
+- Generate embedding vectors ```python speaker_encoder/compute_embeddings.py --use_cuda true /model/path/best_model.pth model/config/path/config.json dataset/path/ output_path``` . This code parses all .wav files at the given dataset path and generates the same folder structure under the output path with the generated embedding files.
+- Watch training on Tensorboard as in TTS