app.py

import gradio as gr
import torch
from transformers import pipeline

username = "AescF"  ## Complete your username
model_id = "AescF/distilhubert-finetuned-gtzan"
device = "cuda:0" if torch.cuda.is_available() else "cpu"
pipe = pipeline("audio-classification", model=model_id, device=device)

# def predict_trunc(filepath):
#     preprocessed = pipe.preprocess(filepath)
#     truncated = pipe.feature_extractor.pad(preprocessed,truncation=True, max_length = 16_000*30)
#     model_outputs = pipe.forward(truncated)
#     outputs = pipe.postprocess(model_outputs)

#     return outputs


def classify_audio(filepath):
    """
      Goes from
      [{'score': 0.8339303731918335, 'label': 'country'},
    {'score': 0.11914275586605072, 'label': 'rock'},]
     to
     {"country":  0.8339303731918335, "rock":0.11914275586605072}
    """
    start_time = timer()
    preds = pipe(filepath)
    # preds = predict_trunc(filepath)
    outputs = {}
    pred_time = round(timer() - start_time, 5)
    for p in preds:
        outputs[p["label"]] = p["score"]
    return outputs


title = "🎵 Music Genre Classifier"
description = """
Demo for a music genre classifier trained on [GTZAN](https://huggingface.co/datasets/marsyas/gtzan)
For more info checkout [GITHUB](https://github.com/AEscF)
"""

filenames = ['blues.00098.wav', "disco.00020.wav", "classical.00075.wav","keyboard-153960.mp3"]
filenames = [[f"./{f}"] for f in filenames]
demo = gr.Interface(
    fn=classify_audio,
    inputs=gr.Audio(type="filepath"),
    outputs=[gr.Label(label="Predictions"), gr.Number(label="Prediction time (s)")],
    title=title,
    description=description,
    examples=filenames,
)
demo.launch()