Music Transcription Helper
Some tools are very usefull whe you need to transcribe music, but some of that can have requirements problems, so I create a docker-compose file to avoid to need that requirements on your computer. It can be founded in this repo.
I will take for example to want to transcribe the flute part of this song that is present on youtube.
Clone repo and build dockers, it will need much times:
git clone https://gitlab.com/nicolalandro/music-transcription-docker-compose-helper.git
docker compose build
Youtube-dl is a tool to download from youtube:
docker-compose run youtube-dl --extract-audio --audio-format mp3 "https://www.youtube.com/watch?v=DPniE-OR9hs"
Demucs is a tool that allow you to split audio in defferente sources wav that represent different instruments, but it does not support the flute so it will be founded into other.wav
docker-compose run demucs -n mdx_extra_q -d cpu "Lo spirito del Signore-DPniE-OR9hs.mp3" -o .
Now is enougth to transcribe manually.
Spleeter is an alternative of demucs, generally perform a bit less well but It have a features that confuse flutes with vocals, so I try to use it to separate flute into other without success.
docker-compose run spleeter separate -p spleeter:2stems -o output "/output/mdx_extra_q/Lo spirito del Signore-DPniE-OR9hs/other.wav"
Omnizart can trascribe the partiture or the chords, in this case I use it for trascribe all music, but tipically this task is not perform good as vocal or instrument solo transcriptions.
docker-compose run omnizart music transcribe "mdx_extra_q/Lo spirito del Signore-DPniE-OR9hs/other.wav"
We are mooving throw much help on music trascription but that AI models are not perfect, but also the fuction of subdivide audio from different surces type is a very usefull feature that help a lot.
I remember that all that works are open source, so we can fill the gap togheter by contributeing!