added tts to whole app.py
Browse files
app.py
CHANGED
@@ -1,49 +1,268 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import torch
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
from transformers import pipeline
|
4 |
|
5 |
-
|
6 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
|
8 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
9 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
pipe = pipeline("text-to-speech", model="kakao-enterprise/vits-ljs", device=device)
|
11 |
|
12 |
-
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
-
|
16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
visible=True)
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
-
|
35 |
-
with gr.Blocks(css=css) as demo_blocks:
|
36 |
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
|
42 |
-
|
43 |
-
|
44 |
-
|
|
|
|
|
|
|
|
|
45 |
|
46 |
-
|
|
|
|
|
|
|
|
|
|
|
47 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
48 |
|
49 |
-
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import numpy as np
|
3 |
+
import json
|
4 |
+
import librosa
|
5 |
+
import os
|
6 |
+
import soundfile as sf
|
7 |
+
import tempfile
|
8 |
+
import uuid
|
9 |
+
import transformers
|
10 |
import torch
|
11 |
+
import time
|
12 |
+
import spaces
|
13 |
+
|
14 |
+
from nemo.collections.asr.models import ASRModel
|
15 |
+
|
16 |
+
from transformers import AutoModelForCausalLM
|
17 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
18 |
|
19 |
from transformers import pipeline
|
20 |
|
21 |
+
# Set an environment variable
|
22 |
+
HF_TOKEN = os.environ.get("HF_TOKEN", None)
|
23 |
+
|
24 |
+
|
25 |
+
SAMPLE_RATE = 16000 # Hz
|
26 |
+
MAX_AUDIO_SECONDS = 40 # wont try to transcribe if longer than this
|
27 |
+
DESCRIPTION = '''
|
28 |
+
<div>
|
29 |
+
<h1 style='text-align: center'>MyAlexa: Voice Chat Assistant</h1>
|
30 |
+
<p style='text-align: center'>MyAlexa is a demo of a voice chat assistant with chat logs that accepts audio input and outputs an AI response. </p>
|
31 |
+
<p>This space uses <a href="https://huggingface.co/nvidia/canary-1b"><b>NVIDIA Canary 1B</b></a> for Automatic Speech-to-text Recognition (ASR), <a href="https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct"><b>Meta Llama 3 8B Insruct</b></a> for the large language model (LLM) and <a href="https://huggingface.co/kakao-enterprise/vits-ljs"><b>VITS-ljs by Kakao Enterprise</b></a> for text to speech (TTS).</p>
|
32 |
+
<p>This demo accepts audio inputs not more than 40 seconds long.</p>
|
33 |
+
<p>Transcription and responses are limited to the English language.</p>
|
34 |
+
</div>
|
35 |
+
'''
|
36 |
+
PLACEHOLDER = """
|
37 |
+
<div style="padding: 30px; text-align: center; display: flex; flex-direction: column; align-items: center;">
|
38 |
+
<img src="./MyAlexaLogo.png" style="width: 80%; max-width: 550px; height: auto; opacity: 0.55; ">
|
39 |
+
<p style="font-size: 28px; margin-bottom: 2px; opacity: 0.65;">What's on your mind?</p>
|
40 |
+
</div>
|
41 |
+
"""
|
42 |
+
# PLACEHOLDER = """
|
43 |
+
# <div style="padding: 30px; text-align: center; display: flex; flex-direction: column; align-items: center;">
|
44 |
+
# <img src="https://i.ibb.co/S35q17Q/My-Alexa-Logo.png" style="width: 80%; max-width: 550px; height: auto; opacity: 0.55; ">
|
45 |
+
# <p style="font-size: 28px; margin-bottom: 2px; opacity: 0.65;">What's on your mind?</p>
|
46 |
+
# </div>
|
47 |
+
# """
|
48 |
|
49 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
50 |
|
51 |
+
### ASR model
|
52 |
+
canary_model = ASRModel.from_pretrained("nvidia/canary-1b").to(device)
|
53 |
+
canary_model.eval()
|
54 |
+
# make sure beam size always 1 for consistency
|
55 |
+
canary_model.change_decoding_strategy(None)
|
56 |
+
decoding_cfg = canary_model.cfg.decoding
|
57 |
+
decoding_cfg.beam.beam_size = 1
|
58 |
+
canary_model.change_decoding_strategy(decoding_cfg)
|
59 |
+
|
60 |
+
### LLM model
|
61 |
+
# Load the tokenizer and model
|
62 |
+
llm_tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct")
|
63 |
+
llama3_model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct", device_map="auto") # to("cuda:0")
|
64 |
+
|
65 |
+
if llm_tokenizer.pad_token is None:
|
66 |
+
llm_tokenizer.pad_token = llm_tokenizer.eos_token
|
67 |
+
|
68 |
+
terminators = [
|
69 |
+
llm_tokenizer.eos_token_id,
|
70 |
+
llm_tokenizer.convert_tokens_to_ids("<|eot_id|>")
|
71 |
+
]
|
72 |
+
|
73 |
+
### TTS model
|
74 |
pipe = pipeline("text-to-speech", model="kakao-enterprise/vits-ljs", device=device)
|
75 |
|
76 |
+
def convert_audio(audio_filepath, tmpdir, utt_id):
|
77 |
+
"""
|
78 |
+
Convert all files to monochannel 16 kHz wav files.
|
79 |
+
Do not convert and raise error if audio is too long.
|
80 |
+
Returns output filename and duration.
|
81 |
+
"""
|
82 |
+
|
83 |
+
data, sr = librosa.load(audio_filepath, sr=None, mono=True)
|
84 |
+
|
85 |
+
duration = librosa.get_duration(y=data, sr=sr)
|
86 |
+
|
87 |
+
if duration > MAX_AUDIO_SECONDS:
|
88 |
+
raise gr.Error(
|
89 |
+
f"This demo can transcribe up to {MAX_AUDIO_SECONDS} seconds of audio. "
|
90 |
+
"If you wish, you may trim the audio using the Audio viewer in Step 1 "
|
91 |
+
"(click on the scissors icon to start trimming audio)."
|
92 |
+
)
|
93 |
+
|
94 |
+
if sr != SAMPLE_RATE:
|
95 |
+
data = librosa.resample(data, orig_sr=sr, target_sr=SAMPLE_RATE)
|
96 |
+
|
97 |
+
out_filename = os.path.join(tmpdir, utt_id + '.wav')
|
98 |
+
|
99 |
+
# save output audio
|
100 |
+
sf.write(out_filename, data, SAMPLE_RATE)
|
101 |
+
|
102 |
+
return out_filename, duration
|
103 |
+
|
104 |
+
def transcribe(audio_filepath):
|
105 |
+
"""
|
106 |
+
Transcribes a converted audio file.
|
107 |
+
Set to english language with punctuations.
|
108 |
+
Returns the transcribed text as a string.
|
109 |
+
"""
|
110 |
|
111 |
+
if audio_filepath is None:
|
112 |
+
raise gr.Error("Please provide some input audio: either upload an audio file or use the microphone")
|
113 |
+
|
114 |
+
utt_id = uuid.uuid4()
|
115 |
+
with tempfile.TemporaryDirectory() as tmpdir:
|
116 |
+
converted_audio_filepath, duration = convert_audio(audio_filepath, tmpdir, str(utt_id))
|
117 |
+
|
118 |
+
# make manifest file and save
|
119 |
+
manifest_data = {
|
120 |
+
"audio_filepath": converted_audio_filepath,
|
121 |
+
"source_lang": "en",
|
122 |
+
"target_lang": "en",
|
123 |
+
"taskname": "asr",
|
124 |
+
"pnc": "yes",
|
125 |
+
"answer": "predict",
|
126 |
+
"duration": str(duration),
|
127 |
+
}
|
128 |
+
|
129 |
+
manifest_filepath = os.path.join(tmpdir, f'{utt_id}.json')
|
130 |
+
|
131 |
+
with open(manifest_filepath, 'w') as fout:
|
132 |
+
line = json.dumps(manifest_data)
|
133 |
+
fout.write(line + '\n')
|
134 |
+
|
135 |
+
# call transcribe, passing in manifest filepath
|
136 |
+
output_text = canary_model.transcribe(manifest_filepath)[0]
|
137 |
+
|
138 |
+
return output_text
|
139 |
+
|
140 |
+
def add_message(history, message):
|
141 |
+
"""
|
142 |
+
Adds the input message in the chatbot.
|
143 |
+
Returns the updated chatbot history.
|
144 |
+
"""
|
145 |
+
history.append((message, None))
|
146 |
+
return history
|
147 |
+
|
148 |
+
def bot(history, message):
|
149 |
+
"""
|
150 |
+
Gets the bot's response and places the user and bot messages in the chatbot
|
151 |
+
Returns the appended chatbot history.
|
152 |
+
"""
|
153 |
+
response = bot_response(message, history)
|
154 |
+
lines = response.split("\n")
|
155 |
+
complete_lines = '\n'.join(lines[2:])
|
156 |
+
answer = ""
|
157 |
+
for character in complete_lines:
|
158 |
+
answer += character
|
159 |
+
new_tuple = list(history[-1])
|
160 |
+
new_tuple[1] = answer
|
161 |
+
history[-1] = tuple(new_tuple)
|
162 |
+
time.sleep(0.05)
|
163 |
+
yield history
|
164 |
+
#return history
|
165 |
+
|
166 |
+
@spaces.GPU()
|
167 |
+
def bot_response(message, history):
|
168 |
+
"""
|
169 |
+
Generates a streaming response using the llama3-8b model.
|
170 |
+
Set max_new_tokens = 100, temperature=0.6, and top_p=0.9
|
171 |
+
Returns the generated response in string format.
|
172 |
+
"""
|
173 |
+
conversation = []
|
174 |
+
for user, assistant in history:
|
175 |
+
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
176 |
+
conversation.append({"role": "user", "content": message})
|
177 |
+
|
178 |
+
input_ids = llm_tokenizer.apply_chat_template(conversation, return_tensors="pt").to(llama3_model.device)
|
179 |
+
|
180 |
+
outputs = llama3_model.generate(
|
181 |
+
input_ids,
|
182 |
+
max_new_tokens = 100,
|
183 |
+
eos_token_id = terminators,
|
184 |
+
do_sample=True,
|
185 |
+
temperature=0.6,
|
186 |
+
top_p=0.9,
|
187 |
+
pad_token_id=llm_tokenizer.pad_token_id,
|
188 |
+
)
|
189 |
+
|
190 |
+
out = outputs[0][input_ids.shape[-1]:]
|
191 |
+
|
192 |
+
return llm_tokenizer.decode(out, skip_special_tokens=True)
|
193 |
+
|
194 |
+
|
195 |
+
def voice_player(history):
|
196 |
+
"""
|
197 |
+
Plays the generated response using the VITS-ljs model.
|
198 |
+
Returns the audio player with the generated response.
|
199 |
+
"""
|
200 |
+
_, text = history[-1]
|
201 |
+
voice = pipe(text)
|
202 |
+
voice = gr.Audio(value = (voice["sampling_rate"], voice["audio"].squeeze()), type="numpy", autoplay=True, label="MyAlexa Response", show_label=True,
|
203 |
visible=True)
|
204 |
+
return voice
|
205 |
+
|
206 |
+
|
207 |
+
with gr.Blocks(
|
208 |
+
title="MyAlexa",
|
209 |
+
css="""
|
210 |
+
textarea { font-size: 18px;}
|
211 |
+
""",
|
212 |
+
theme=gr.themes.Default(text_size=gr.themes.sizes.text_lg) # make text slightly bigger (default is text_md )
|
213 |
+
) as demo:
|
214 |
+
|
215 |
+
gr.HTML(DESCRIPTION)
|
216 |
+
chatbot = gr.Chatbot(
|
217 |
+
[],
|
218 |
+
elem_id="chatbot",
|
219 |
+
bubble_full_width=False,
|
220 |
+
placeholder=PLACEHOLDER,
|
221 |
+
label='MyAlexa'
|
222 |
+
)
|
223 |
+
with gr.Row():
|
224 |
+
with gr.Column():
|
225 |
+
gr.HTML(
|
226 |
+
"<p><b>Step 1:</b> Upload an audio file or record with your microphone.</p>"
|
227 |
+
)
|
228 |
+
|
229 |
+
audio_file = gr.Audio(sources=["microphone", "upload"], type="filepath")
|
230 |
+
|
231 |
+
|
232 |
+
with gr.Column():
|
233 |
|
234 |
+
gr.HTML("<p><b>Step 2:</b> Submit your recorded or uploaded audio as input and wait for MyAlexa's response.</p>")
|
|
|
235 |
|
236 |
+
submit_button = gr.Button(
|
237 |
+
value="Submit audio",
|
238 |
+
variant="primary"
|
239 |
+
)
|
240 |
|
241 |
+
chat_input = gr.Textbox(
|
242 |
+
label="Transcribed text:",
|
243 |
+
interactive=False,
|
244 |
+
placeholder="Transcribed text will appear here.",
|
245 |
+
elem_id="chat_input",
|
246 |
+
visible=True # set to True to see processing time of asr transcription
|
247 |
+
)
|
248 |
|
249 |
+
out_audio = gr.Audio(
|
250 |
+
value = None,
|
251 |
+
label="Response Voice Player",
|
252 |
+
show_label=True,
|
253 |
+
visible=True # set to True to see processing time of tts audio generation
|
254 |
+
)
|
255 |
|
256 |
+
chat_msg = chat_input.change(add_message, [chatbot, chat_input], [chatbot], api_name="add_message_in_chatbot")
|
257 |
+
bot_msg = chat_msg.then(bot, [chatbot, chat_input], chatbot, api_name="bot_response_in_chatbot")
|
258 |
+
voice_msg = bot_msg.then(voice_player, chatbot, out_audio, api_name="bot_response_voice_player")
|
259 |
+
|
260 |
+
submit_button.click(
|
261 |
+
fn=transcribe,
|
262 |
+
inputs = [audio_file],
|
263 |
+
outputs = [chat_input]
|
264 |
+
)
|
265 |
|
266 |
+
demo.queue()
|
267 |
+
if __name__ == "__main__":
|
268 |
+
demo.launch()
|