Skip to content

keunwoochoi/dl4mir

Folders and files

NameName
Last commit message
Last commit date
Sep 21, 2017
Jan 1, 2018
Jan 1, 2018
Jul 15, 2017
Jul 15, 2017
Dec 31, 2017
Feb 6, 2019
Apr 14, 2018
Jul 15, 2017
Apr 14, 2018
Jul 15, 2017
Apr 14, 2018
Jul 15, 2017
Oct 3, 2017
May 24, 2017
Oct 8, 2024
Apr 14, 2018
May 25, 2017

Repository files navigation

dl4mir: A Tutorial on Deep Learning for MIR

by Keunwoo Choi ([email protected])

This is a repo for my tutorial paper; A Tutorial on Deep Learning for Music Information Retrieval.

Tutorials

  1. Example 1: Pitch detector with a dense layer
  2. Example 2: Chord recogniser with a convnet
  3. Example 3: Setup config.json
  4. Example 4: download and preprocess
  5. Real examples with real datasets! * Example 5-1: Time-varying classification example using Jamendo dataset * Example 5-2: Time-invariant classification example using FMA dataset

Prerequisites

$ pip install -r requirements.txt
$ git clone https://github.com/keunwoochoi/kapre.git
$ cd kapre
$ python setup.py install

to install

  • Librosa, Keras, Numpy, Matplotlib, Future
  • kapre

Notes

  • Datasets is removed from Kapre and the codes are directly imported into here.

Datasets

Dataset management

  • GTZan: (30s, 10 genres, 1,000 mp3)
  • MagnaTagATune: (29s, 188 tags, 25,880 mp3) for tagging and triplet similarity
  • MusicNet: (full length 330 classicals music, note-wise annotations)
  • FMA: small/medium/large/full collections, up to 100+K songs from free music archieve, for genre classification. With genre hierarchy, pre-computed features, splits, etc.
  • Jamendo: 61/16/24 songs for vocal activity detection

Some links

  • Repo
  • Slides
    • Deep Neural Networks in MIR: A tutorial focusing on feature learning, beat/rhythm analysis, structure analysis. Also a nice literature overview including publications by year, conference, task, network types, input representations, frame work, etc. By Meinard Muller et al.
    • DL in music informatics: ISMIR 2014 tutorial.
  • Documents, books
    • Deep learning book: The first deep learning textbook. by Ian Goodfellow and Yoshua Bengio and Aaron Courville.
  • Online

Cite?

@article{choi2017tutorial,
  title={A Tutorial on Deep Learning for Music Information Retrieval},
  author={Choi, Keunwoo and Fazekas, Gy{\"o}rgy and Cho, Kyunghyun and Sandler, Mark},
  journal={arXiv:1709.04396},
  year={2017}
}

Or visit the paper page on Google scholar for potential updates.