app.py

import streamlit as st
import base64
import io
import time
from huggingface_hub import InferenceClient
from gtts import gTTS
from pydub import AudioSegment
import speech_recognition as sr
import streamlit_mic_recorder as smr

pre_prompt_text = ""

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

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("Recognized text:")
            st.write(audio_text)
            st.success("Voice Recognized.")
    except sr.UnknownValueError:
        st.warning("The audio could not be recognized. Did you try to record something?")
        audio_text = ""
    except sr.RequestError:
        st.error("Push/Talk to start!")
        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]"
        st.session_state.pre_prompt_sent = True

    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
    temperature = max(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):
    tts = gTTS(text=text, lang='es')
    audio_fp = io.BytesIO()
    tts.write_to_fp(audio_fp)
    audio_fp.seek(0)
    return audio_fp

def main():
    st.write("Di la palabra XAMAN para empezar o DETENTE para procesar")

    recording = st.empty()
    st.write(smr.audio_recorder())

    if st.button("Procesar"):
        audio_data = smr.get_audio()
        st.audio(audio_data, format="audio/wav")

        audio_text = recognize_speech(audio_data)
        if audio_text:
            output, audio_file = generate(audio_text)
            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()