Professionally-produced music recordings
This year’s edition features many novelties:
- New dataset and Python tools to handle it
- 100 training songs, 50 test songs, all email@example.com kHz produced recordings
- All songs include drums, bass, other, vocals stems
- Songs are encoded in the Native Instruments stems format, with a tool to convert them back and forth to wav and to load them directly in Python.
- Automatic download of data
- Python code to analyze results and produce plots for your own paper.
The purpose of this task is to evaluate source separation algorithms for estimating one or more sources from a set of mixtures in the context of professionally-produced music recordings.
The data set consists of a total of 150 full-track songs of different styles and includes both the stereo mixtures and the original sources, divided between a training subset and a test subset (see Section I below).
The participants are kindly asked to download the data set and evaluation packages, run the evaluation function and follow the instructions for submission of results presented in section II below.
I. The Dataset
For information about the dataset and get started, visit the sigsep MUS 2018 data webpage.
This year, the dataset considered for the MUS task is called SigSep MUS 2018 database.
In total, the dataset comprises 100 songs in the train set, and 50 songs in the test set.
Warning!Note: the sigsep MUS 2018 test set is different from DSD100 test set. We took into account relevant complaints from participants that the same artist could be in both train and test sets. This does not happen anymore.
II. The Evaluation
Visit the MUS 2018 submission page for information on how to submit your results.
Evaluation for SiSEC MUS 2018 has been simplified compared to previous years, and enjoys much automation.
In case you have any questions on the SiSEC 2018 MUS task, feel free to ask Fabian Stöter (fabian-robert.stoter[at]inria.fr)