4 ms·
Show HN: Destructive Commons : Voice / Instrumental automatic separation
What is Destructive Commons?
Demixing and sampling. You send the mix as mp3 file, we run a complicated algorithm, and you get separated background and lead tracks as mp3 files. With those, you can do sampling or automatic karaoke.
?
Destructive Commons is a wonderful piece of signal processing technology which automatically unmixes an mp3 music file into its vocal and instrumental parts. You upload a mp3 file, you leave your email, and a few minutes later, you get a link for your separated tracks, which is up for a few hours.
Does it work?
Depends. Basically, the algorithm separates stuff that is repetitive from stuff that is not. Usually, you get vocals and instrumentals as separated tracks. You may end up with some guitar solo or such stuff. Indeed, those are not repetitive neither. Furthermore, separation is not perfect at all and you usually end up with sound artifacts. However, with sufficiently noisy covers or hardcore drums over them, that should be nice. Results depend of the original song, the weather and your cat.
- quematech 12y agoWe hope you enjoy it (we're terribly proud of the design....)
- diydsp 12y agoOK! I got my results back! That was fun! On the rock song I sent it, it kind of worked! In general, it separated the higher frequencies of the vocals better than the lower. And in general, the instrumental track is higher quality than the vocal-only track! Also, in the quieter parts of the song - no guitar, just drums and vocals, the hi-hats came through on the vocal side. The separation worked best when there was singing and guitars and drums. It's a bizarre, fun effect to play both tracks back, vocals panned left, instruments panned right. Some of the singing pans from left to right as it slips in and out of the classification :) Thanks for this! If you ever felt like describing how it works more, it would be fascinating to hear!
- quematech 12y agoHi. Glad you enjoyed it. We use some advanced signal processing algorithms. Basically, the algorithm separates stuff that is repetitive from stuff that is not, which explains the erratic result. For example, if the instrumental is very repetitive, you'll have quite good results, as with some commercial stuff. But you can get sometimes easily a solo with this... We'll be releasing a new version in two monthes which will give users access to other parameters, like the period used to define the "repetitive" part.
- diydsp 12y agoCool, this looks like a lot of fun! I'm preparing my youtube->mp4->mp3 chain right now so I can try it out on Jets to Brazil's "Chinatown." (It goes youtube-dl.exe then ffmpeg, but I hated the shifty windows 7zip program b/c it tried to install a new browser, so I'm using Cygwin's 7zip to decompress windows ffmpeg and cygwin is updating everything under the sun, including emacs...). BTW, the reason I want to do this is to transcribe the vocal part of the song into sheet music. This will work way better when the vocals are separated. I'll let you know how it goes, thank you.