|
--- |
|
language: |
|
- en |
|
tags: |
|
- JETS |
|
- LightSpeech |
|
- MB-MelGAN |
|
- Text-to-Speech |
|
datasets: |
|
- CMUArctic |
|
- Hi-Fi |
|
pipeline_tag: text-to-speech |
|
--- |
|
|
|
# TTS Models |
|
|
|
Here you can find models compatible with [Balacoon software](https://balacoon.com). |
|
There are several model types to be aware of: |
|
|
|
* `*_jets_cpu.addon` - JETS models for synthesis on CPU, compatible with |
|
[balacoon_tts](https://gemfury.com/balacoon/python:balacoon-tts) python package. |
|
A tutorial on how to use it can be found [here](https://balacoon.com/use/tts/package). |
|
Those are high-end models, producing 24khz audio. |
|
* `*_light_cpu.addon` - Light models for lightning-fast synthesis on-device. |
|
Usage is the same as of JETS models, but the naturalness of synthesized audio is compromised |
|
in favor of speed. Models produce 16khz audio. |
|
* `*_jets_gpu.addon` - JETS models for synthesis on GPU, compatible with |
|
[balacoon/tts_server](https://hub.docker.com/r/balacoon/tts_server) docker image. |
|
A tutorial on how to use it can be found [here](https://balacoon.com/use/tts/service). |
|
Exactly the same as jets_cpu models, but repacked for GPU. |
|
|
|
You can check the interactive demo |
|
[balacoon/tts](https://huggingface.co/spaces/balacoon/tts) space. |
|
|
|
# List of available models |
|
|
|
- en-US locale |
|
- [CMUArtic databases](http://festvox.org/cmu_arctic/) |
|
- <mark>en_us_cmuartic_jets_cpu.addon</mark> |
|
- <mark>en_us_cmuartic_jets_gpu.addon</mark> |
|
- [Hi-Fi audiobooks dataset](https://arxiv.org/abs/2104.01497) |
|
- <mark>en_us_hifi_jets_cpu.addon</mark> |
|
- <mark>en_us_hifi92_light_cpu.addon</mark> - trained only on "92" speaker |