Skip to content
Pronounced as "musician", musicnn is a set of pre-trained deep convolutional neural networks for music audio tagging.
Jupyter Notebook Python
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
audio Notebook functioning with basic demo Jul 8, 2019
images Debugging functions, and renaming tagging_example.ipynb Jul 31, 2019
musicnn Update extractor.py Oct 15, 2019
.gitignore Adding new MSD model and VGGs Jul 26, 2019
DOCUMENTATION.md Update DOCUMENTATION.md Aug 1, 2019
FAQs.md Update FAQs.md Sep 17, 2019
LICENSE.md Minor fixes for pushing to PyPI Jul 18, 2019
MANIFEST.in Update MANIFEST.in Aug 18, 2019
README.md Update README.md Aug 1, 2019
musicnn_example.ipynb Updating notebooks Jul 30, 2019
setup.py
tagging_example.ipynb
vgg_example.ipynb Updating notebooks Jul 30, 2019

README.md

musicnn

Pronounced as "musician", musicnn is a set of pre-trained musically motivated convolutional neural networks for music audio tagging. This repository also includes some pre-trained vgg-like baselines.

Check the documentation and our basic / advanced examples to understand how to use musicnn.

Do you have questions? Check the FAQs.

Installation

pip install musicnn

or, to get bigger models and all the documentation (including jupyter notebooks), install from source:

git clone https://github.com/jordipons/musicnn.git

python setup.py install

Predict tags

From within python, you can estimate the topN tags:

from musicnn.tagger import top_tags
top_tags('./audio/joram-moments_of_clarity-08-solipsism-59-88.mp3', model='MTT_musicnn', topN=10)

['techno', 'electronic', 'synth', 'fast', 'beat', 'drums', 'no vocals', 'no vocal', 'dance', 'ambient']

Let's try another song!

top_tags('./audio/TRWJAZW128F42760DD_test.mp3')

['guitar', 'piano', 'fast']

From the command-line, you can also print the topN tags on the screen:

python -m musicnn.tagger file_name.ogg --print
python -m musicnn.tagger file_name.au --model 'MSD_musicnn' --topN 3 --length 3 --overlap 1.5 --print

or save to a file:

python -m musicnn.tagger file_name.wav --save out.tags
python -m musicnn.tagger file_name.mp3 --model 'MTT_musicnn' --topN 10 --length 3 --overlap 1 --print --save out.tags

Extract the Taggram

You can also compute the taggram using python (see our basic example for more details on how to depict it):

from musicnn.extractor import extractor
taggram, tags = extractor('./audio/joram-moments_of_clarity-08-solipsism-59-88.mp3', model='MTT_musicnn')

Taggram

The above analyzed music clips are included in the ./audio/ folder of this repository.

You can listen to those and evaluate musicnn yourself!

You can’t perform that action at this time.