app.py

import streamlit as st
import base64
import io
from huggingface_hub import InferenceClient
from gtts import gTTS
from audiorecorder import audiorecorder
import speech_recognition as sr
from pydub import AudioSegment

if "history" not in st.session_state:
    st.session_state.history = []

def recognize_speech(audio_data, show_messages=True):
    recognizer = sr.Recognizer()
    audio_recording = sr.AudioFile(audio_data)

    with audio_recording as source:
        audio = recognizer.record(source)

    try:
        audio_text = recognizer.recognize_google(audio, language="es-ES")
        if show_messages:
            st.subheader("Texto Reconocido:")
            st.write(audio_text)
            st.success("Reconocimiento de voz completado.")
    except sr.UnknownValueError:
        st.warning("No se pudo reconocer el audio. ¿Intentaste grabar algo?")
        audio_text = ""
    except sr.RequestError:
        st.error("Hablame para comenzar!")
        audio_text = ""

    return audio_text

def format_prompt(message, history):
    prompt = "<s>"

    for user_prompt, bot_response in history:
        prompt += f"[INST] {user_prompt} [/INST]"
        prompt += f" {bot_response}</s> "

    prompt += f"[INST] {message} [/INST]"
    return prompt

def generate(audio_text, history, temperature=None, max_new_tokens=512, top_p=0.95, repetition_penalty=1.0):
    client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")

    temperature = float(temperature) if temperature is not None else 0.9
    if temperature < 1e-2:
        temperature = 1e-2
    top_p = float(top_p)

    generate_kwargs = dict(
        temperature=temperature,
        max_new_tokens=max_new_tokens,
        top_p=top_p,
        repetition_penalty=repetition_penalty,
        do_sample=True,
        seed=42,
    )

    formatted_prompt = format_prompt(audio_text, history)
    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
    response = ""

    for response_token in stream:
        response += response_token.token.text
    
    response = ' '.join(response.split()).replace('</s>', '')
    audio_file = text_to_speech(response, speed=1.3)
    return response, audio_file

def text_to_speech(text, speed=1.3):
    tts = gTTS(text=text, lang='es')
    audio_fp = io.BytesIO()
    tts.write_to_fp(audio_fp)
    audio_fp.seek(0)
    audio = AudioSegment.from_file(audio_fp, format="mp3")
    modified_speed_audio = audio.speedup(playback_speed=speed)
    modified_audio_fp = io.BytesIO()
    modified_speed_audio.export(modified_audio_fp, format="mp3")
    modified_audio_fp.seek(0)
    return modified_audio_fp

def main():
    st.title("Chatbot de Voz a Voz")
    audio_data = audiorecorder("Habla para grabar", "Deteniendo la grabación...")

    if not audio_data.empty():
        st.audio(audio_data.export().read(), format="audio/wav")
        audio_data.export("audio.wav", format="wav")
        audio_text = recognize_speech("audio.wav")

        if audio_text:
            output, audio_file = generate(audio_text, history=st.session_state.history)  

            if audio_text:  
                st.session_state.history.append((audio_text, output))  

            if audio_file is not None:
                st.markdown(
                    f"""
                    <audio autoplay="autoplay" controls="controls" src="data:audio/mp3;base64,{base64.b64encode(audio_file.read()).decode()}" type="audio/mp3" id="audio_player"></audio>
                    """,
                    unsafe_allow_html=True
                )

if __name__ == "__main__":
    main()