Skip to content

Latest commit

 

History

History
134 lines (79 loc) · 7.24 KB

README.md

File metadata and controls

134 lines (79 loc) · 7.24 KB
Raw

DJtransGAN: Automatic DJ Transitions with Differentiable Audio Effects and Generative Adversarial Networks (Data Generation Pipeline)

This repository contains the code for "Automatic DJ Transitions with Differentiable Audio Effects and Generative Adversarial Networks" 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2022) Bo-Yu Chen, Wei-Han Hsu, Wei-Hsiang Liao, Marco A. Martínez-Ramírez, Yuki Mitsufuji, Yi-Hsuan Yang

Overview

This repo contain the data generation pipeline code of DJtransGAN which is the essential part of DJtransGAN. After generaing the dataset, please refer to DJtransGAN repo to train or use the model. The completed pipeline contain three part to generate mixable pair

  1. Feature extraction: beat & downbeat tracking, key estimation and structure boundary detection.
  2. Mixability estimation: use muscial rule and neural network to find the two mixabile pair.
  3. Alignment: bpm, key and cue region aligment to make sure to segment can mix perfectly.

and a independent script to generate the mix segment.

Furthermore, if you want to hear more audio example, please check our demo page here.

Dataset

We collected two datasets to train our proposed approaches: DJ mixset from Livetracklist and individual EDM tracks from MTG-Jamendo-Dataset.

To be more specific, We in-house collect long DJ mixsets from Livetracklist and only consider the mixset with mix tag to ensure the quality of the mixset. Furthermore, we select the individual EDM track from MTG-Jamendo-Dataset, which means the track with an EDM tag in the total collection. The detailed information about using these two datasets to train our model is described in Section 3.1; please check it if you are interested.

Unfortunately, we can not provide our training dataset for reproducing the results because of license issues. However, we release the training code and pre-trained model for you to try. Contact me or open the pull request if you have any other issues.

Setup

Install


pip install -r requirements.txt

Set up extenal package

We use music puzzle game to conduct the mixability estimation, thus you need to clone and rename it to music_puzzle_games in the begining.


git clone https://github.com/remyhuang/music-puzzle-games.git music_puzzle_games

Configuration

Next, you should set the configuration in pipeline/config/settings.py for global usage of the repo, Most important of all, you should set the path of TRACK_DIR, MIX_DIR.

  1. TRACK_DIR : the directory conatin the collection of EDM tracks, you should put all your EDM track under the {TRACK_DIR}/audio.
  2. MIX_DIR : the directory conatin the collection of mix and its cue point, you should put all you mix under the {MIX_DIR}/audio and provide a meta data file {MIX_DIR}/meta.json include the cue point (only need cue out point as known as switch point) of individual mix (we provide a sample in the repo for you to check the format).

Usage

We release several usage examples in examples/ and script/ for data generation and the usage of invidual step in pipeline. please check it, if you want to use or modify it.

Mixable pair generation

To generate the mixable pair, you need to run two scripts sequentially. First, you should run the script in script/create_segment.py, which is going to extract the music segment by segmenting the collection of EDM tracks.

python create_segment.py [--feature=(bool, ex: 1)] [--stem=(bool, ex: 1)] [--segment=(bool, ex:1)] [--n_core=(int, ex: 5)] [--n_gpu=(int, ex: 0)]
  • --stem : specify whether to cache the source separtion result for instrument detection in the begining to speed up the training.
  • --feature : specify whether to cache the feature extraction result in the begining to speed up the training.
  • --n_core : the number of mutlti-processor you want to use.
  • --n_gpu : speicify which gpu you want to use.

Next, you need to run the script in script/create_pair.py to match the music segment which is sutible to mix together.

python create_pair.py [--match=(str, e.g: None, all, nn, rule)] [--n_sample=(int, ex: 10)] [--n_core=(int, ex: 4)]
  • --match : speicify the mixabiltiy estimaiton approach.
    • None: random pick one segment as matching result.
    • rule: use musical rule to filter out the data and random pick one segment as matching result.
    • nn: use neural network (SEN form music puzzle game) to pick one segemnt as matching result.
    • all: use musical rule to filter out the data and use neural network to pick one segment as matching result.
  • --n_core : the number of mutlti-processor you want to use.
  • --n_sample : the maximum number of sample for mixability estimation to speed up the training.

Mix generation

To train the DJtransGAN, we still need a professional DJ mix as a reference for GAN to learn. Thus, you can run the script in script/create_mix.py to get a such data. Please remember to provide essential material to MIX_DIR.

python create_mix.py [--n_core=(int, ex: 4)]
  • --n_core : the number of mutlti-processor you want to use.

Citation

If you use any of our code in your work please consider citing us.

  @inproceedings{chen2022djtransgan,
    title={Automatic DJ Transitions with Differentiable Audio Effects and Generative Adversarial Networks},
    author={Chen, B. Y., Hsu, W. H., Liao, W. H., Ramírez, M. A. M., Mitsufuji, Y., & Yang, Y. H.},
    booktitle={ICASSP},
    year={2022}}

Acknowledgement

This repo is done during the internship in the Sony Group Corporation with outstanding mentoring by my incredible mentors in Sony Wei-Hsiang Liao, Marco A. Martínez-Ramírez, and Yuki Mitsufuji and my colleague Wei-Han Hsu and advisor Yi-Hsuan Yang in Academia Sinica. The results are a joint effort with Sony Group Corporation and Academia Sinica. I sincerely appreciate all the support made by them to make this research happen. Moreover, please check the other excellent AI research made by Sony here and their recent work "FxNorm-automix" and "distortionremoval" which is going to present in ISMIR 2022.

License

Copyright © 2022 Bo-Yu Chen

Licensed under the MIT License (the "License"). You may not use this package except in compliance with the License. You may obtain a copy of the License at

https://opensource.org/licenses/MIT

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.