Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -1,46 +1,354 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
2 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
3 |
-
import
|
4 |
-
from f5_infer import F5TTS
|
5 |
|
6 |
-
|
7 |
-
|
8 |
-
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
-
#
|
11 |
-
|
|
|
|
|
|
|
|
|
12 |
|
13 |
-
#
|
14 |
-
|
15 |
-
|
|
|
|
|
|
|
16 |
|
17 |
-
#
|
18 |
-
|
19 |
|
20 |
-
#
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
-
|
43 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
|
45 |
-
|
46 |
-
|
|
|
1 |
+
# ruff: noqa: E402
|
2 |
+
# Above allows ruff to ignore E402: module level import not at top of file
|
3 |
+
|
4 |
+
import re
|
5 |
+
import tempfile
|
6 |
+
|
7 |
import gradio as gr
|
8 |
+
import numpy as np
|
9 |
+
import soundfile as sf
|
10 |
+
import torchaudio
|
11 |
+
from cached_path import cached_path
|
12 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
13 |
+
import torch
|
|
|
14 |
|
15 |
+
from f5_tts.model import DiT
|
16 |
+
from f5_tts.infer.utils_infer import (
|
17 |
+
load_vocoder,
|
18 |
+
load_model,
|
19 |
+
preprocess_ref_audio_text,
|
20 |
+
infer_process,
|
21 |
+
remove_silence_for_generated_wav,
|
22 |
+
)
|
23 |
|
24 |
+
# Intentar importar 'spaces' para determinar si se est谩 usando Hugging Face Spaces
|
25 |
+
try:
|
26 |
+
import spaces
|
27 |
+
USING_SPACES = True
|
28 |
+
except ImportError:
|
29 |
+
USING_SPACES = False
|
30 |
|
31 |
+
# Decorador para utilizar GPU si est谩 disponible
|
32 |
+
def gpu_decorator(func):
|
33 |
+
if USING_SPACES:
|
34 |
+
return spaces.GPU(func)
|
35 |
+
else:
|
36 |
+
return func
|
37 |
|
38 |
+
# Cargar el vocoder
|
39 |
+
vocoder = load_vocoder()
|
40 |
|
41 |
+
# Cargar el modelo F5-TTS
|
42 |
+
F5TTS_model_cfg = dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)
|
43 |
+
F5TTS_ema_model = load_model(
|
44 |
+
DiT, F5TTS_model_cfg, str(cached_path("hf://jpgallegoar/F5-Spanish/model_1200000.safetensors"))
|
45 |
+
)
|
46 |
+
|
47 |
+
# Variables globales para el modelo de chat
|
48 |
+
chat_model_state = None
|
49 |
+
chat_tokenizer_state = None
|
50 |
+
|
51 |
+
@gpu_decorator
|
52 |
+
def generate_response(messages, model, tokenizer):
|
53 |
+
"""Genera una respuesta usando el modelo de chat"""
|
54 |
+
text = tokenizer.apply_chat_template(
|
55 |
+
messages,
|
56 |
+
tokenize=False,
|
57 |
+
add_generation_prompt=True,
|
58 |
+
)
|
59 |
+
|
60 |
+
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
61 |
+
generated_ids = model.generate(
|
62 |
+
**model_inputs,
|
63 |
+
max_new_tokens=512,
|
64 |
+
temperature=0.7,
|
65 |
+
top_p=0.95,
|
66 |
+
)
|
67 |
+
|
68 |
+
generated_ids = [
|
69 |
+
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
70 |
+
]
|
71 |
+
return tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
72 |
+
|
73 |
+
def traducir_numero_a_texto(texto):
|
74 |
+
"""Convierte n煤meros en texto a su representaci贸n en palabras en espa帽ol"""
|
75 |
+
texto_separado = re.sub(r'([A-Za-z])(\d)', r'\1 \2', texto)
|
76 |
+
texto_separado = re.sub(r'(\d)([A-Za-z])', r'\1 \2', texto_separado)
|
77 |
|
78 |
+
def reemplazar_numero(match):
|
79 |
+
numero = match.group()
|
80 |
+
return num2words(int(numero), lang='es')
|
81 |
+
|
82 |
+
texto_traducido = re.sub(r'\b\d+\b', reemplazar_numero, texto_separado)
|
83 |
+
|
84 |
+
return texto_traducido
|
85 |
+
|
86 |
+
@gpu_decorator
|
87 |
+
def infer(
|
88 |
+
ref_audio_orig, ref_text, gen_text, model, remove_silence, cross_fade_duration=0.15, speed=1, show_info=gr.Info
|
89 |
+
):
|
90 |
+
"""Genera el audio sintetizado a partir del texto"""
|
91 |
+
ref_audio, ref_text = preprocess_ref_audio_text(ref_audio_orig, ref_text, show_info=show_info)
|
92 |
+
|
93 |
+
ema_model = F5TTS_ema_model
|
94 |
+
|
95 |
+
if not gen_text.startswith(" "):
|
96 |
+
gen_text = " " + gen_text
|
97 |
+
if not gen_text.endswith(". "):
|
98 |
+
gen_text += ". "
|
99 |
+
|
100 |
+
gen_text = gen_text.lower()
|
101 |
+
gen_text = traducir_numero_a_texto(gen_text)
|
102 |
+
|
103 |
+
final_wave, final_sample_rate, combined_spectrogram = infer_process(
|
104 |
+
ref_audio,
|
105 |
+
ref_text,
|
106 |
+
gen_text,
|
107 |
+
ema_model,
|
108 |
+
vocoder,
|
109 |
+
cross_fade_duration=cross_fade_duration,
|
110 |
+
speed=speed,
|
111 |
+
show_info=show_info,
|
112 |
+
progress=gr.Progress(),
|
113 |
+
)
|
114 |
+
|
115 |
+
# Eliminar silencios si est谩 activado
|
116 |
+
if remove_silence:
|
117 |
+
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
|
118 |
+
sf.write(f.name, final_wave, final_sample_rate)
|
119 |
+
remove_silence_for_generated_wav(f.name)
|
120 |
+
final_wave, _ = torchaudio.load(f.name)
|
121 |
+
final_wave = final_wave.squeeze().cpu().numpy()
|
122 |
+
|
123 |
+
return (final_sample_rate, final_wave)
|
124 |
+
|
125 |
+
def load_chat_model():
|
126 |
+
"""Carga el modelo de chat y el tokenizer"""
|
127 |
+
global chat_model_state, chat_tokenizer_state
|
128 |
+
if chat_model_state is None:
|
129 |
+
model_name = "Qwen/Qwen2.5-3B-Instruct"
|
130 |
+
chat_model_state = AutoModelForCausalLM.from_pretrained(
|
131 |
+
model_name, torch_dtype=torch.float16, device_map="auto"
|
132 |
+
)
|
133 |
+
chat_tokenizer_state = AutoTokenizer.from_pretrained(model_name)
|
134 |
+
return chat_model_state, chat_tokenizer_state
|
135 |
+
|
136 |
+
with gr.Blocks() as app_chat:
|
137 |
+
gr.Markdown(
|
138 |
+
"""
|
139 |
+
# Chat de Voz
|
140 |
+
隆Mant茅n una conversaci贸n con una IA usando tu voz de referencia!
|
141 |
+
1. Sube un clip de audio de referencia y opcionalmente su transcripci贸n.
|
142 |
+
2. Carga el modelo de chat.
|
143 |
+
3. Graba tu mensaje a trav茅s de tu micr贸fono.
|
144 |
+
4. La IA responder谩 usando la voz de referencia.
|
145 |
+
"""
|
146 |
+
)
|
147 |
+
|
148 |
+
if not USING_SPACES:
|
149 |
+
load_chat_model_btn = gr.Button("Cargar Modelo de Chat", variant="primary")
|
150 |
+
|
151 |
+
chat_interface_container = gr.Column(visible=False)
|
152 |
+
|
153 |
+
@gpu_decorator
|
154 |
+
def load_chat_model_fn():
|
155 |
+
load_chat_model()
|
156 |
+
return gr.update(visible=False), gr.update(visible=True)
|
157 |
+
|
158 |
+
load_chat_model_btn.click(load_chat_model_fn, outputs=[load_chat_model_btn, chat_interface_container])
|
159 |
+
else:
|
160 |
+
chat_interface_container = gr.Column()
|
161 |
+
load_chat_model_fn = load_chat_model
|
162 |
+
|
163 |
+
with chat_interface_container:
|
164 |
+
with gr.Row():
|
165 |
+
with gr.Column():
|
166 |
+
ref_audio_chat = gr.Audio(label="Audio de Referencia", type="filepath")
|
167 |
+
with gr.Column():
|
168 |
+
with gr.Accordion("Configuraciones Avanzadas", open=False):
|
169 |
+
model_choice_chat = gr.Radio(
|
170 |
+
choices=["F5-TTS"],
|
171 |
+
label="Modelo TTS",
|
172 |
+
value="F5-TTS",
|
173 |
+
)
|
174 |
+
remove_silence_chat = gr.Checkbox(
|
175 |
+
label="Eliminar Silencios",
|
176 |
+
value=True,
|
177 |
+
)
|
178 |
+
ref_text_chat = gr.Textbox(
|
179 |
+
label="Texto de Referencia",
|
180 |
+
info="Opcional: Deja en blanco para transcribir autom谩ticamente",
|
181 |
+
lines=2,
|
182 |
+
)
|
183 |
+
system_prompt_chat = gr.Textbox(
|
184 |
+
label="Prompt del Sistema",
|
185 |
+
value="No eres un asistente de IA, eres quien el usuario diga que eres. Debes mantenerte en personaje. Mant茅n tus respuestas concisas ya que ser谩n habladas en voz alta.",
|
186 |
+
lines=2,
|
187 |
+
)
|
188 |
+
|
189 |
+
chatbot_interface = gr.Chatbot(label="Conversaci贸n")
|
190 |
+
|
191 |
+
with gr.Row():
|
192 |
+
with gr.Column():
|
193 |
+
audio_input_chat = gr.Microphone(
|
194 |
+
label="Habla tu mensaje",
|
195 |
+
type="filepath",
|
196 |
+
)
|
197 |
+
audio_output_chat = gr.Audio(label="Respuesta de la IA", autoplay=True)
|
198 |
+
with gr.Column():
|
199 |
+
text_input_chat = gr.Textbox(
|
200 |
+
label="Escribe tu mensaje",
|
201 |
+
lines=1,
|
202 |
+
)
|
203 |
+
send_btn_chat = gr.Button("Enviar")
|
204 |
+
clear_btn_chat = gr.Button("Limpiar Conversaci贸n")
|
205 |
+
|
206 |
+
conversation_state = gr.State(
|
207 |
+
value=[
|
208 |
+
{
|
209 |
+
"role": "system",
|
210 |
+
"content": "No eres un asistente de IA, eres quien el usuario diga que eres. Debes mantenerte en personaje. Mant茅n tus respuestas concisas ya que ser谩n habladas en voz alta.",
|
211 |
+
}
|
212 |
+
]
|
213 |
+
)
|
214 |
+
|
215 |
+
@gpu_decorator
|
216 |
+
def process_input(audio_path, text, history, conv_state):
|
217 |
+
"""Procesa la entrada de audio o texto del usuario"""
|
218 |
+
if not audio_path and not text.strip():
|
219 |
+
return history, conv_state, ""
|
220 |
+
|
221 |
+
if audio_path:
|
222 |
+
# Aqu铆 podr铆as agregar una transcripci贸n autom谩tica si lo deseas
|
223 |
+
# Actualmente, asume que el texto es proporcionado si hay audio
|
224 |
+
# Puedes integrar Whisper u otro modelo de transcripci贸n si es necesario
|
225 |
+
pass
|
226 |
+
|
227 |
+
if not text.strip():
|
228 |
+
return history, conv_state, ""
|
229 |
+
|
230 |
+
conv_state.append({"role": "user", "content": text})
|
231 |
+
history.append((text, None))
|
232 |
+
|
233 |
+
response = generate_response(conv_state, chat_model_state, chat_tokenizer_state)
|
234 |
+
|
235 |
+
conv_state.append({"role": "assistant", "content": response})
|
236 |
+
history[-1] = (text, response)
|
237 |
+
|
238 |
+
return history, conv_state, response
|
239 |
+
|
240 |
+
@gpu_decorator
|
241 |
+
def generate_audio_response(response, ref_audio, ref_text, model, remove_silence):
|
242 |
+
"""Genera el audio de respuesta para la IA"""
|
243 |
+
if not response or not ref_audio:
|
244 |
+
return None
|
245 |
+
|
246 |
+
audio_result, _ = infer(
|
247 |
+
ref_audio,
|
248 |
+
ref_text,
|
249 |
+
response,
|
250 |
+
model,
|
251 |
+
remove_silence,
|
252 |
+
cross_fade_duration=0.15,
|
253 |
+
speed=1.0,
|
254 |
+
show_info=print,
|
255 |
+
)
|
256 |
+
return audio_result
|
257 |
+
|
258 |
+
def clear_conversation_fn():
|
259 |
+
"""Limpia la conversaci贸n"""
|
260 |
+
return [], [
|
261 |
+
{
|
262 |
+
"role": "system",
|
263 |
+
"content": "No eres un asistente de IA, eres quien el usuario diga que eres. Debes mantenerte en personaje. Mant茅n tus respuestas concisas ya que ser谩n habladas en voz alta.",
|
264 |
+
}
|
265 |
+
]
|
266 |
+
|
267 |
+
def update_system_prompt_fn(new_prompt):
|
268 |
+
"""Actualiza el prompt del sistema y reinicia la conversaci贸n"""
|
269 |
+
new_conv_state = [{"role": "system", "content": new_prompt}]
|
270 |
+
return [], new_conv_state
|
271 |
+
|
272 |
+
# Manejar la entrada de audio
|
273 |
+
audio_input_chat.stop_recording(
|
274 |
+
process_input,
|
275 |
+
inputs=[audio_input_chat, text_input_chat, chatbot_interface, conversation_state],
|
276 |
+
outputs=[chatbot_interface, conversation_state, text_input_chat],
|
277 |
+
).then(
|
278 |
+
generate_audio_response,
|
279 |
+
inputs=[text_input_chat, ref_audio_chat, ref_text_chat, model_choice_chat, remove_silence_chat],
|
280 |
+
outputs=[audio_output_chat],
|
281 |
+
).then(
|
282 |
+
lambda: None,
|
283 |
+
None,
|
284 |
+
audio_input_chat,
|
285 |
+
)
|
286 |
+
|
287 |
+
# Manejar la entrada de texto
|
288 |
+
text_input_chat.submit(
|
289 |
+
process_input,
|
290 |
+
inputs=[audio_input_chat, text_input_chat, chatbot_interface, conversation_state],
|
291 |
+
outputs=[chatbot_interface, conversation_state, text_input_chat],
|
292 |
+
).then(
|
293 |
+
generate_audio_response,
|
294 |
+
inputs=[text_input_chat, ref_audio_chat, ref_text_chat, model_choice_chat, remove_silence_chat],
|
295 |
+
outputs=[audio_output_chat],
|
296 |
+
).then(
|
297 |
+
lambda: None,
|
298 |
+
None,
|
299 |
+
text_input_chat,
|
300 |
+
)
|
301 |
+
|
302 |
+
# Manejar el bot贸n de enviar
|
303 |
+
send_btn_chat.click(
|
304 |
+
process_input,
|
305 |
+
inputs=[audio_input_chat, text_input_chat, chatbot_interface, conversation_state],
|
306 |
+
outputs=[chatbot_interface, conversation_state, text_input_chat],
|
307 |
+
).then(
|
308 |
+
generate_audio_response,
|
309 |
+
inputs=[text_input_chat, ref_audio_chat, ref_text_chat, model_choice_chat, remove_silence_chat],
|
310 |
+
outputs=[audio_output_chat],
|
311 |
+
).then(
|
312 |
+
lambda: None,
|
313 |
+
None,
|
314 |
+
text_input_chat,
|
315 |
+
)
|
316 |
+
|
317 |
+
# Manejar el bot贸n de limpiar conversaci贸n
|
318 |
+
clear_btn_chat.click(
|
319 |
+
clear_conversation_fn,
|
320 |
+
outputs=[chatbot_interface, conversation_state],
|
321 |
+
)
|
322 |
+
|
323 |
+
# Manejar cambios en el prompt del sistema
|
324 |
+
system_prompt_chat.change(
|
325 |
+
update_system_prompt_fn,
|
326 |
+
inputs=system_prompt_chat,
|
327 |
+
outputs=[chatbot_interface, conversation_state],
|
328 |
+
)
|
329 |
+
|
330 |
+
def main():
|
331 |
+
if not USING_SPACES:
|
332 |
+
import click
|
333 |
+
|
334 |
+
@click.command()
|
335 |
+
@click.option("--port", "-p", default=None, type=int, help="Puerto para ejecutar la aplicaci贸n")
|
336 |
+
@click.option("--host", "-H", default=None, help="Host para ejecutar la aplicaci贸n")
|
337 |
+
@click.option(
|
338 |
+
"--share",
|
339 |
+
"-s",
|
340 |
+
default=False,
|
341 |
+
is_flag=True,
|
342 |
+
help="Compartir la aplicaci贸n a trav茅s de un enlace compartido de Gradio",
|
343 |
+
)
|
344 |
+
@click.option("--api", "-a", default=True, is_flag=True, help="Permitir acceso a la API")
|
345 |
+
def run_app(port, host, share, api):
|
346 |
+
print("Iniciando la aplicaci贸n de Chat AI...")
|
347 |
+
app_chat.queue(api_open=api).launch(server_name=host, server_port=port, share=share, show_api=api)
|
348 |
+
|
349 |
+
run_app()
|
350 |
+
else:
|
351 |
+
app_chat.queue().launch()
|
352 |
|
353 |
+
if __name__ == "__main__":
|
354 |
+
main()
|