app.py

import io
import base64
import numpy as np
import soundfile as sf
from gtts import gTTS
import streamlit as st
import speech_recognition as sr
from huggingface_hub import InferenceClient
from streamlit_mic_recorder import mic_recorder

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

if "pre_prompt_sent" not in st.session_state:
    st.session_state.pre_prompt_sent = False

pre_prompt_text = "eres una IA conductual, tus respuestas serán breves."

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

    try:
        adjusted_audio_data = sf.resample(audio_data, sample_rate, 16000, subtype='PCM_16')
        audio_text = recognizer.recognize_google(adjusted_audio_data, 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>"

    if not st.session_state.pre_prompt_sent:
        prompt += f"[INST]{pre_prompt_text}[/INST]"

    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)
    return audio_fp

def audio_play(audio_fp):
    st.audio(audio_fp.read(), format="audio/mp3", start_time=0)

def display_recognition_result(audio_text, output, audio_file):
    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)

def main():
    if not st.session_state.pre_prompt_sent:
        st.session_state.pre_prompt_sent = True

    audio = mic_recorder(start_prompt="▶️", stop_prompt="🛑", key='recorder')

    if audio:
        st.audio(audio['bytes'], format="audio/wav")
        audio_bytes = np.frombuffer(audio["bytes"], dtype=np.int16)
        sample_rate = audio["sample_rate"]

        audio_text = recognize_speech(audio_bytes, sample_rate)
        
        if audio_text:
            st.session_state.history.append((audio_text, ""))

if __name__ == "__main__":
    main()