The Sound Demixing Challenge 2023 (MDX23) is an opportunity for researchers and machine learning enthusiasts to test their skills by creating a system to perform audio source separation.
Given an audio signal as input (referred to as a “mixture”), you must decompose in its different parts across a few categories.
This year’s SDX Challenge includes two new tracks: Music Demixing Track and Cinematic Sound Demixing Track.
Music Demixing Track
This task will focus on music source separation. Participants will submit systems that separate a song into four instruments: vocals, bass, drums, and so on. It focuses on the music source separation of an audio signal into four instrument tracks.
Cinematic Sound Demixing Track
This task focuses on the source separation of a cinematic audio track into three sub-tracks: “dialogue”, “sound effects”, and “music”.
This challenge features a prize pool of 42,000 USD across two tracks and five leaderboards.
- Music Demixing Track: 32,000 USD
- Cinematic Sound Separation Track: 10,000 USD
The warmup round for Music Demixing Track is now live!
Check out the dataset, starter kit and make your first submission.
