app.py

import gradio as gr
from audioseal import AudioSeal
import torch
import numpy as np
import traceback

def detect_watermark(audio_data, sample_rate):
    try:
        audio_array, _ = audio_data
        if audio_array.ndim == 1:
            audio_array = np.expand_dims(audio_array, axis=0)
        waveform = torch.tensor(audio_array, dtype=torch.float32)
        if waveform.ndim == 2:
            waveform = waveform.unsqueeze(0)
        detector = AudioSeal.load_detector("audioseal_detector_16bits")
        result, message = detector.detect_watermark(waveform, message_threshold=0.5)
        detection_result = "AI-generated" if result else "genuine"
        return f"This audio is likely {detection_result} based on watermark detection."
    except Exception as e:
        error_traceback = traceback.format_exc()
        return f"Error: {str(e)}\n{error_traceback}"

interface = gr.Interface(fn=detect_watermark,
                         inputs=[gr.Audio(label="Upload your audio", type="numpy"),
                                 gr.Number(label="Sample Rate", value=44100, visible=False)],
                         outputs="text")

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