app.py

import gradio as gr
from transformers import Speech2TextProcessor, Speech2TextForConditionalGeneration
import librosa
import torch
import os
from huggingface_hub import login

# Obtener el token desde las variables de entorno
token = os.getenv("HF_TOKEN")
if token:
    login(token=token)
else:
    raise ValueError("El token de Hugging Face no está configurado en las variables de entorno.")

device = "cuda" if torch.cuda.is_available() else "cpu"

# Cargar el modelo
repo_name = "HugoZeballos/rapa_nui_asr_2"  # Ajusta al nombre de tu modelo en Hugging Face
processor = Speech2TextProcessor.from_pretrained(repo_name)
model = Speech2TextForConditionalGeneration.from_pretrained(repo_name).to(device)

# Cambiar `source` a una configuración válida o eliminarlo
inputs = gr.Audio(type="filepath")
outputs = gr.Textbox(label="Transcripción")

def transcribe(audio_path):
    audio, sr = librosa.load(audio_path, sr=16000)
    inputs = processor(audio, sampling_rate=sr, return_tensors="pt", padding="longest").to("cuda")

    with torch.no_grad():
        predicted_ids = model.generate(inputs["input_features"], attention_mask=inputs["attention_mask"])
    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
    return transcription
    
# Crear interfaz
interface = gr.Interface(
    fn=transcribe,
    inputs=inputs,
    outputs=outputs,
    title="ASR Demo"
)

# Ejecutar la app
if __name__ == "__main__":
    interface.launch()