app.py

import gradio as gr
import torch
import torchaudio
import librosa
import numpy as np
from transformers import Wav2Vec2Processor, Wav2Vec2ForSequenceClassification
import torch.nn.functional as F
import torchaudio.transforms as T

model_name = "Mahmoud59/wav2vec2-fake-audio-detector"
processor = Wav2Vec2Processor.from_pretrained(model_name)
model = Wav2Vec2ForSequenceClassification.from_pretrained(model_name)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)


def classify_audio(audio_path):
    # Load audio with librosa
    waveform, sample_rate = librosa.load(audio_path, sr=16000)  # librosa automatically resamples to 16kHz

    # Convert to tensor
    waveform = torch.tensor(waveform, dtype=torch.float32)

    # Process audio
    inputs = processor(waveform, sampling_rate=16000, return_tensors="pt", padding=True)
    inputs = {key: val.to(device) for key, val in inputs.items()}

    # Run inference
    with torch.no_grad():
        outputs = model(**inputs)

    # Get prediction
    pred = torch.argmax(outputs.logits, dim=-1).item()
    return "Fake" if pred == 1 else "Real"

# 6) Build Gradio interface
demo = gr.Interface(
    fn=classify_audio,
    inputs=gr.Audio(type="filepath"),
    outputs="text",
    title="Wav2Vec2 Deepfake Detection",
    description="Upload an audio sample to check if it is fake or real, along with confidence."
)

if __name__ == "__main__":
    demo.launch()