/drumsep

Hybrid Demucs model for drum separation

Primary LanguageShell

drumsep

Drum separation based on a Hybrid Demucs model.
Code to facilitate access to the model presented in the thesis "Separación de fuentes en grabaciones de batería mediante aprendizaje automático" (2022).

Universidad Nacional de Quilmes.

The model takes as input an audio file (mp3, wav, flac, or ogg) corresponding to a drum recording and exports 4 audio files classified as:

  • Bombo (Kick)
  • Redoblante (Snare)
  • Platillos (Cymbals)
  • Toms

With Google Colab

To easily use the model online:
drumsep Notebook

With Linux

Having pip installed.

  1. Clone repo.
  2. Install.
bash drumsepInstall
  1. Separate
    <PATH_IN> Path to the audio file or directory containing audio files to be separated.
    <PATH_OUT> Path to the directory where the generated audio files will be exported. 
bash drumsep "<PATH_IN>" "<PATH_OUT>"

(Efforts are currently underway to advance research and document progress for this project, with the ultimate objective of sharing valuable insights with the wider community).