Audio Signal Separation addresses the problem of segregating certain signals from an audio mixture. Focusing on the separation of musical audio signals, we addressed this problem from different angles in the context of several public and industrial research projects. Listen to some demo sounds in DEMOS tab.

Orchestral music signals separation

We investigate the separation of instrumental signals from orchestra recordings. In this task we can consider both blind and score-informed situations, as well as different type of mixture signals, from a stereo downmix to raw multi-microphone signals captured on a concert hall.

This research is conducted as part of the PHENICX project (2013-2016).

Lead vocals, bass and percussion separation

In the musical context, we focused on the analysis and extraction of the predominant voice from polyphonic music (ex: Marxer et al., 2012), and percussion components (e.g. Janer et al. 2012). These algorithms have various applications including musical production (e.g remixes), entertainment (e.g. karaoke) or cultural heritage (e.g. restoration).

This research was conducted as part of a joint-research collaboration with Yamaha Corp. Japan (2009-2012).


In the scope of our research on audio Source eSeparation, we generated datasets that we release publicly in order to be used by other researchers.

  • MASS Music Audio Signal Separation dataset (link)
  • DREANSS DRum Event ANnotations for Source Separation (link)


Past projects

  • Audio Scanner:  the aim of this project was to manually assist realtime musical audio signal separation in stereo recordings by using panning, phase alignment and other magnitude cues in a STFT scheme. [More info]


For inquiries related to our research on audio signal separation please send an email to Jordi Janer (jordi . janer at upf . edu)