Skip to content
Tools to convert sigsep mus dataset from STEMS <-> WAV
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

SIGSEP MUS STEMS Encoder/Decoder

The purpose of this tool is to decode and encode the sigsep mus stems dataset. By providing a docker based installation setup, we can make sure that all users deal with the same decoded PCM data and therefore foster reproducible research.


The easiest way to set up the encoder and decoder for the MUS dataset is to install docker.

Pull Docker Container

Pull our precompiled sigsep-mus-io image from dockerhub:

docker pull faroit/sigsep-mus-io

Alternatively (or if you want to customize the Dockerfile) you can clone the repository and build the docker container from command line using:

docker build -t mus .


Decode SIGSEP MUS to WAV format

Unzip the MUS dataset to any location of your choice. Lets assume you have it unpacked in ~/data/MUS/.

Using Bash on Mac/Linux

Run the decoding scripts with the following command below. Make sure to change the volume mounts -v ABSOLUTE/PATH/TO/YOUR/MUS/FOLDER:/data in the docker command.

docker run --rm -v $(~)/data/MUS:/data faroit/sigsep-mus-io /scripts/

Using Windows (10)

In Windows Command Line (cmd) you can use the directory variables to mount the data directories. Assuming that the data lies in the MUS subfolder of your home directory, you may use:

docker run --rm -v %HOMEDRIVE%"%HOMEPATH%/MUS":/data faroit/sigsep-mus-io /scripts/

Encode SIGSEP MUS to STEMS format

To encode a MUS estimates directory, you can use the sh script. Again, if you have saved the data directoy in a different location than the default folder, please change it

docker run --rm -v $(~)/data/ESTIMATE_DIR:/data faroit/sigsep-mus-io /mus/

Manual Conversion


  • FFMPEG 3.4
  • MP4Box 7.1
  • gnu-parallel


Run the or scripts manually.

You can’t perform that action at this time.