Music Puzzle Games

TensorFlow implementation of Generating Music Medleys via Playing Music Puzzle Games

• Similarity Embedding Network (SEN) trained on downbeat-informed data

Please cite this paper if this code/work is helpful:

@inproceedings{huang2018generating,
  title={Generating music medleys via playing music puzzle games},
  author={Huang, Yu-Siang and Chou, Szu-Yu and Yang, Yi-Hsuan},
  booktitle={AAAI},
  pages={2281--2288},
  year={2018}
}

Environment

• Python 3.6
• TensorFlow 1.2.0
• NumPy 1.14.3
• LibROSA 0.6.2
• Pydub 0.22.1

There are already three sample audio clips in the data folder. You can directly run the code.

$ git clone https://github.com/remyhuang/music-puzzle-games.git
$ cd music-puzzle-games
$ python main.py

Use

Replace the sample audio files (mp3 format) in the data for your own purpose. (Noted: Due to using the brute-force method to find the best permutation, the total number of audio files should not exceed 11.)

Output

Three default output files:

• best_permutation.txt: the best permutation calculated by pairwise similarities
• output.csv: pairwise similarities
• output.mp3: concatenated audio file of the best permutation

Contact

Please feel free to contact Yu-Siang Huang if you have any questions.