It can be tough to know what we would like to listen when faced with an archive of thousand of musics. For our project we wondered how we could automatically generate a list of songs that would correspond to one's taste using the Free Music Archive avalaible on the internet.

Since the dataset is quite large, to use these scripts you have to download the following data: fma_metadata.zip (342 MiB). Extract the content of fma_metadata folder into the data folder.

• Generate the corresponding small FMA dataset graph by running Graph creator.ipynb. (You only need to do it one time)
• Put your music in the audio folder, you can add as many songs as you wish, each subfolder corresponds to one desired playlist. Two Demos folder are included as examples.
• Run Extractor.ipynb to extract the features of each of your songs. (It will take some time)
• Finally, open Playlist creator.ipynb. Modify the parameters as you like.

Note: If you afterwards add other songs to be analyzed, you only need to re-execute Extractor.ipynb and Playlist creator.ipynb

• You may want to use the provided environment.yml conda environment to execute all the scripts properly
• If you have any trouble with Extractor.ipynb, already-made demo lists of extracted song features are avalaible in the data folder, you can simply run Playlist creator.ipynb

FMA: A Dataset For Music Analysis: Michaël Defferrard, Kirell Benzi, Pierre Vandergheynst, Xavier Bresson, 2017.
GitHub. paper.

• Team: 30
• Students: Joachim Tapparel, Tim Tuuva, Lucas Biotto, Anael Buchegger
• Dataset: Free Music Archive