app.py

import numpy as np
import torch
import librosa
import gradio as gr
from transformers import AutoModelForAudioClassification, Wav2Vec2Processor
import logging

logging.basicConfig(level=logging.INFO)

model_path = "./"
model = AutoModelForAudioClassification.from_pretrained(model_path)
processor = Wav2Vec2Processor.from_pretrained(model_path)

def preprocess_audio(audio_path, sr=16000):
    audio, _ = librosa.load(audio_path, sr=sr)
    audio, _ = librosa.effects.trim(audio)
    return audio

def extract_features(audio, sr=16000):
    inputs = processor(audio, sampling_rate=sr, return_tensors="pt", padding=True)
    return inputs

def predict_voice(audio_file_path):
    try:
        audio = preprocess_audio(audio_file_path)
        features = extract_features(audio)
        
        with torch.no_grad():
            outputs = model(**features)
            logits = outputs.logits
            predicted_index = logits.argmax(dim=-1)
            label = processor.decode(predicted_index)
            confidence = torch.softmax(logits, dim=-1).max().item() * 100
        
        result = f"The voice is classified as '{label}' with a confidence of {confidence:.2f}%."
        logging.info("Prediction successful.")
    except Exception as e:
        result = f"Error during processing: {e}"
        logging.error(result)

    return result

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