andreasjansson / musicseg_deepemb

Code for paper: "Deep Embeddings and Section Fusion Improve Music Segmentation"

Deep Embeddings and Section Fusion Improve Music Segmentation

Installation

Step 1: install non-python dependencies

Install non-python dependencies:

• ffmpeg
• sox (with support for mp3)
• libsndfile

On Debian/Ubuntu the command would be:

apt-get update --fix-missing && apt-get install libsndfile1 ffmpeg libsox-fmt-all sox -y

Step 2: install the musicsections package

cd musicsections
pip install -e .

Example

import musicsections

deepsim_model_folder = "/path/to/deepsim/model/folder"
fewshot_model_folder = "/path/to/fewshot/model/folder"

model_deepsim = musicsections.load_deepsim_model(deepsim_model_folder)
model_fewshot = musicsections.load_fewshot_model(fewshot_model_folder)

audiofile = "/path/to/audiofile.mp3"

segmentations, features = musicsections.segment_file(
    audiofile, 
    deepsim_model=model_deepsim,
    fewshot_model=model_fewshot,
    min_duration=8,
    mu=0.5,
    gamma=0.5,
    beats_alg="madmom",
    beats_file=None)

musicsections.plot_segmentation(segmentations)

Reference

@inproceedings{Salamon:Segmentation:ISMIR:2021,
	Author = {J. Salamon and O. Nieto and N.J. Bryan},
	Booktitle = {Proc.~22nd International Conference on Music Information Retrieval (ISMIR)},
	Month = {Nov.},
	Title = {Deep Embeddings and Section Fusion Improve Music Segmentation},
	Year = {2021}}