app.py

import gradio as gr
import tensorflow as tf
import librosa
import numpy as np
import tempfile

# Diccionario de etiquetas
labels = ['down', 'go', 'left', 'no', 'off', 'on', 'right', 'stop', 'up', 'yes']

def extract_features(file_name):
    try:
        audio, sample_rate = librosa.load(file_name, res_type='kaiser_fast') 
        mfccs = librosa.feature.mfcc(y=audio, sr=sample_rate, n_mfcc=40)
        mfccsscaled = np.mean(mfccs.T,axis=0)
        
    except Exception as e:
        print(f"Error encountered while parsing file: {file_name}")
        print(e)  # Imprime la excepción completa
        return None 
     
    return mfccsscaled

def classify_audio(audio_file):
    print(f"Tipo de audio_file: {type(audio_file)}") 

    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_file:
        tmp_file.write(audio_file.read())
        tmp_file_path = tmp_file.name

    # Preprocesa el audio (con extract_features())
    features = extract_features(tmp_file_path)

    # Si features es None, hubo un error en extract_features
    if features is None:
        return "Error al procesar el audio" 
    
    features = features.reshape(1, -1) # Redimensiona a (1, 40)
    
    # Si tu modelo necesita 3 dimensiones, redimensiona a (1, 40, 1)
    # features = features.reshape(1, 40, 1)

    # Carga del modelo para usar la CPU
    model = tf.keras.models.load_model('my_model.h5', compile=False)

    # Realiza la predicción en la CPU
    with tf.device('/CPU:0'):
        prediction = model.predict(features)
        predicted_label_index = np.argmax(prediction) 
    
    # Devuelve la etiqueta predicha
    predicted_label = labels[predicted_label_index]
    return predicted_label

iface = gr.Interface(
    fn=classify_audio,
    inputs=gr.Audio(type="filepath"),
    outputs="text",
    title="Clasificación de audio simple",
    description="Sube un archivo de audio para clasificarlo."
)

iface.launch()