This repo contains the code for a streamlit-based demo of our project (main repo) on Deep Learning for music genre classification as part of the Numerical Analysis for Machine Learning course at Politecnico di Milano, A.Y 2022-2023.
The demo consists of a streamlit app that allows you to upload a local audio file
(with supported formats those of torchaudio.load(...))
or paste a YouTube URL for the model to infer the music genre. In order to use the app you need to run the following command
streamlit run 01_Music_Genre_Classification.pyor python -m streamlit run 01_Music_Genre_Classification.py.
Once you input an audio, you will be able to hear it back and observe its spectrogram representation.
The model will then run inference on the given audio. The stacked chart represents the probabilities of each genre across time. The overall scores for the top-5 genres are displayed. Lastly, an interactive representation of the model's latent space is displayed. After completion of inference, the representation of the given song in that feature space will be displayed as black dots.
pip install -r requirements.txtIn addition to the libraries listed in requirements.txt, you will need to have FFmpeg
installed on your machine with the CLI commands accessible from the environment where the
app will run. This might require you to copy the ffmpeg.exe and ffprobe.exe executables
in the directory where your python environment's python.exe is located.