app.py

import gradio as gr
from transformers import AutoFeatureExtractor, AutoModelForAudioClassification
import torch
import librosa
import numpy as np

local_model_path = "./"
extractor = AutoFeatureExtractor.from_pretrained(local_model_path)
model = AutoModelForAudioClassification.from_pretrained(local_model_path)

def preprocess_audio(audio_file_path, target_sample_rate=16000):
    # Load the audio file, ensuring mono conversion
    waveform, _ = librosa.load(audio_file_path, sr=target_sample_rate, mono=True)
    # Normalizing waveform to be between -1 and 1
    waveform = librosa.util.normalize(waveform)
    return waveform, target_sample_rate

def predict_voice(audio_file_path):
    try:
        waveform, sample_rate = preprocess_audio(audio_file_path)
        # Ensure waveform is a float32 array
        waveform = waveform.astype(np.float32)
        inputs = extractor(waveform, return_tensors="pt", sampling_rate=sample_rate)
        
        with torch.no_grad():
            outputs = model(**inputs)
        
        logits = outputs.logits
        predicted_index = logits.argmax()
        label = model.config.id2label[predicted_index.item()]
        confidence = torch.softmax(logits, dim=1).max().item() * 100
        
        result = f"The voice is classified as '{label}' with a confidence of {confidence:.2f}%."
    except Exception as e:
        # Improved error handling for debugging
        result = f"Error during processing: {e}"
    
    return result

iface = gr.Interface(
    fn=predict_voice,
    inputs=gr.Audio(label="Upload Audio File", type="filepath"),
    outputs=gr.Textbox(label="Prediction"),
    title="Voice Authenticity Detection",
    description="Detects whether a voice is real or AI-generated. Upload an audio file to see the results."
)

iface.launch()