Is it performant enough to be used real time? I tried somewhat similar stuff to extract the vocals out of songs, I could make a workable demo (not real time though), but my lazy ass hasn't touched it since.
Good question! Someone asked me a little while back on Twitter the same thing. github.com/librosa/librosa/issues/220 Kind of. You can use core.stream for that. librosa.github.io/librosa/generated/librosa.core.stream.html The problem is some of the methods will no doubt cause some latency issues. The Python language itself usually isn't great for time-sensitive applications, unless relying on some external C binary. This might also be worth reading. blog.francoismaillet.com/epic-celebration/
@@bytecauldron I remember watching a video of Brian McFee, demoing librosa, he said that they tried using cython for better performance but the improvements were insignificant. Lemme try librosa.stream, I had however changed my strategy to load songs first, do the hpss thing and then only find the notes of vocals in real time, but again I had to do a custom implementation for the notes, if streaming works, it would become easier.
@@bytecauldron I also found out why am I subscribed to this channel, th-cam.com/video/oz8uUIy8Rlg/w-d-xo.html , it helped me, thanks, lol, sorry for spamming.
I haven't attempted but it looks like it can. librosa.org/librosa_gallery/auto_examples/plot_vocal_separation.html#sphx-glr-auto-examples-plot-vocal-separation-py Spleeter also looks like a promising library to check out. github.com/deezer/spleeter
Omg, good job! Could you do sth associated with sound note recognition?
Great content. Subbed!
Is it performant enough to be used real time? I tried somewhat similar stuff to extract the vocals out of songs, I could make a workable demo (not real time though), but my lazy ass hasn't touched it since.
Good question! Someone asked me a little while back on Twitter the same thing.
github.com/librosa/librosa/issues/220
Kind of. You can use core.stream for that. librosa.github.io/librosa/generated/librosa.core.stream.html
The problem is some of the methods will no doubt cause some latency issues. The Python language itself usually isn't great for time-sensitive applications, unless relying on some external C binary.
This might also be worth reading. blog.francoismaillet.com/epic-celebration/
@@bytecauldron I remember watching a video of Brian McFee, demoing librosa, he said that they tried using cython for better performance but the improvements were insignificant. Lemme try librosa.stream, I had however changed my strategy to load songs first, do the hpss thing and then only find the notes of vocals in real time, but again I had to do a custom implementation for the notes, if streaming works, it would become easier.
@@bytecauldron I was a bit wrong in whatever I said, th-cam.com/video/MhOdbtPhbLU/w-d-xo.html , here is the link, watched it like an year ago
@@bytecauldron I also found out why am I subscribed to this channel, th-cam.com/video/oz8uUIy8Rlg/w-d-xo.html , it helped me, thanks, lol, sorry for spamming.
@@abhay8437 No worries! Happy I was able to help.
How to do most things in python --
import fairy_dust
fairy_dust.get-da-mf-bass(my_sounds_file.wav)
Fucking love python.
Also this is an incredibly well made video
Will it work to saperat vocal to music ?
I haven't attempted but it looks like it can.
librosa.org/librosa_gallery/auto_examples/plot_vocal_separation.html#sphx-glr-auto-examples-plot-vocal-separation-py
Spleeter also looks like a promising library to check out. github.com/deezer/spleeter
nice