• TalkBass has been independent since 1998. Add your voice.
    Create a free account to reply to discussions, view embedded media, and browse with fewer display ads.
    Join free Log in
    Want zero display ads or expanded classifieds tools? Compare plans.

Anybody seen this? (machine learning track separation)

andruca

andruca

Mar 31, 2004
3,422
4,299
52
Madrid (Spain)
www.andresdemarco.info
#1
Just ran into this, thru' a post in some Facebook group. I'm overwhelmed at how it knocked off my bass in this track.

Original track...


Bassless track...
http://andresdemarco.info/intento_andar_sin_bajo.mp3

Amazed at this. Big blow to the backing track "industry" tho'. Only defect is the generated audio needs some 9-10db boost afterwards to reach a volume similar to the original track amplitude.

Moises.ai

It has several "presets" (one is "bassless", the one I used). I'll need to use some more complex separation to make practice tracks that are bot bassless and voiceless (I'm the one singing too, there's no such preset to supress both bass and vocals).

(BTW, not affiliated in any way, just tried it and liked it so I thought I'd share)

EDIT: drumless version (not perfect -can hear some slight beats thru' the bass- but still pretty awesome)... http://andresdemarco.info/intento_andar_sin_bateria.mp3
 
Last edited:
  • Like
Reactions: FaithNoMan and JulienP.
#6
seamonkey said:
One of the other cool things about this is it is open source (MIT license)
deezer/spleeter

I look forward to some Python coders jumping on this on doing some more refining! :)
Click to expand...

I am indeed a Python coder. Just have no experience in audio processing AT ALL (my breadwinner spectrum is automation related -parking/ticketing/toll collection/etc.-), completely lack such knowledge. For years I've been thinking about the divelopment of an "Ouya of multieffects" (an open-source multi-effects unit you can upload new pedals to as XML "engine" files). Only thing I could barely contribute to on such a project is the functional specs (lots of fly hours on the multiple interface paradigms for multieffects, been intensively using MANY brands and generations of them for over 30 years). The idea came about after demoing a Zoom MS-100BT (the one you can upload effect models you get from an app/store via bluetooth -just a huge config file for the audio engine, I guess-). Unfortunately I know zero about digital audio analysis/processing.
 
  • Like
Reactions: seamonkey
#12
I use Izotope RX7 Rebalance, and it is pretty similar. It's generally pretty good at removing a sound, but with limitations. It works better for bass and drums as they tend to be very different from the other sounds. If you have two instruments in the same frequency range, it's harder to separate. Trying to isolate vox from an acoustic guitar is much harder and has lots of artifacts.
 
  • Like
Reactions: andruca and JulienP.
#13
Oh, a system that can remove your bass track?
They should rename it the:

Learning
Audio
Removal
System

upload_2020-1-8_10-54-47.jpeg


That does seem pretty cool, though. I practice by playing along with songs I like and sometimes recording my bass to them. This seems really useful for that.
 
  • Like
Reactions: AlexanderB, andruca and JulienP.

Latest posts

Top Bottom
Back