Last example using Spleeter for source separation, this time using ADD lossless audio from CD.
I was hoping the AI would perform better with a lossless audio file since the artifacts in the split audio remind me a little of the GSM codec.

Conclusion: it doesn't matter, the AI does a great job regardless of audio codec.

Clearly a LOT of work has gone in to Spleeter and it does great work but it seems to have trouble the longer the audio file is.
I mean, the initial 60 seconds is great but then looses quality. not sure if that's just layering of the song or something to do with the AI learning.
It would seem best suited to extracting vocals with drums. For re-sampling purposes this is ideal since you need the drumbeat to lock the beat in applications like Traktor or pro-tools.

In any case, this is going to be a great tool for preparing samples for use in live performances.

Thanks to the Spleeter peoples :)
[ Ссылка ]

свернуть