When rehearsing, alone or with others, I sometimes like to play along to backing tracks with a few of the original instrument parts removed. My hybrid drumming rig works well for this, but preparing the backing tracks can be a pain. Typically, I’d download an MP3 of the song, manually upload it to a stem-separation service, then re-export it in the TM-2’s expected format: stereo 44.1 kHz PCM WAV. This works, but preparing a couple of songs for a rehearsal can take quite a bit of time.
So I built a tool to automate that process. It accepts an audio file or a permitted video source, separates stems through LALAL.AI, then prepares the final track in the TM-2’s expected format.
For example, the following command creates a backing track without vocals or drums:
poetry run python -m track_automation.cli \
"<Youtube_URI>" \
"<Track_Name>" \
--include guitar bass piano
The --include option selects the parts to keep, leaving the vocals and drums to be played live.
This tool has made a fiddly music-preparation job much quicker, and the resulting audio files are surprisingly high quality.