A small, end-to-end project that finetunes a Hugging Face Wav2Vec2 model on the classic GTZAN dataset to classify music genres.

• Self-supervised Wav2Vec2 finetuned for 10 GTZAN genres
• Best validation accuracy: 87%
• Training with transformers.Trainer (HF)
• Gradio web UI to upload an audio file and get predictions

from transformers import pipeline

model_id = "hangnguyen25/wav2vec2-base-finetuned-gtzan"
pipe = pipeline(
    "audio-classification",
    model=model_id
    )

preds = pipe("path/to/your/audio.wav")
print(preds)

• Dataset: GTZAN
• Libraries: Hugging Face transformers, datasets, evaluate, accelerate; librosa; gradio; PyTorch