app.py

import gradio as gr
from transformers import pipeline
from pydub import AudioSegment
from pydub.utils import make_chunks
import os

# Modelo ASR
modelo = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-large-xlsr-53-spanish")

def dividir_audio(input_path, output_dir, chunk_length_ms=30000):
    audio = AudioSegment.from_file(input_path)
    chunks = make_chunks(audio, chunk_length_ms)
    os.makedirs(output_dir, exist_ok=True)

    for i, chunk in enumerate(chunks):
        chunk_name = os.path.join(output_dir, f"chunk_{i}.wav")
        chunk.export(chunk_name, format="wav")

def transcribir_audio(audio_path):
    return modelo(audio_path)["text"]

def transcribir(audio_file):
    output_dir = "chunks"
    dividir_audio(audio_file, output_dir)

    transcripcion_final = []
    for filename in sorted(os.listdir(output_dir)):
        if filename.endswith(".wav"):
            filepath = os.path.join(output_dir, filename)
            transcripcion_final.append(transcribir_audio(filepath))

    return " ".join(transcripcion_final)

with gr.Interface(
    fn=transcribir,
    inputs=gr.Audio(type="filepath", label="Sube tu archivo de audio"),
    outputs="text",
    title="Transcriptor de Audio"
) as interfaz:
    interfaz.launch()