| import gradio as gr | |
| from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2Processor | |
| import torch | |
| import torchaudio | |
| model_name = "Mrkomiljon/voiceGUARD/wav2vec2_finetuned_model" | |
| model = Wav2Vec2ForSequenceClassification.from_pretrained(model_name) | |
| processor = Wav2Vec2Processor.from_pretrained(model_name) | |
| model.eval() | |
| def classify_audio(audio_file): | |
| waveform, sample_rate = torchaudio.load(audio_file) | |
| if sample_rate != 16000: | |
| resampler = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000) | |
| waveform = resampler(waveform) | |
| if waveform.size(1) > 16000 * 10: | |
| waveform = waveform[:, :16000 * 10] | |
| elif waveform.size(1) < 16000 * 10: | |
| waveform = torch.nn.functional.pad(waveform, (0, 16000 * 10 - waveform.size(1))) | |
| if waveform.ndim > 1: | |
| waveform = waveform[0] | |
| inputs = processor(waveform.numpy(), sampling_rate=16000, return_tensors="pt") | |
| with torch.no_grad(): | |
| logits = model(**inputs).logits | |
| predicted_label = logits.argmax(dim=-1).item() | |
| return predicted_label | |
| interface = gr.Interface( | |
| fn=classify_audio, | |
| inputs=gr.Audio(source="upload", type="filepath"), | |
| outputs="label", | |
| title="Audio Classifier", | |
| description="Upload an audio file to classify its label as AI-generated or Real." | |
| ) | |
| interface.launch() | |