4

I am looking for a software that can separate the voice track from the rest of the audio.

It should:

  • take a WAV or MP3 file as input.
  • create 2 WAV or MP3 files as output: one for the music, one for the voice.
  • works offline (no Internet connection needed).
  • is gratis (Open Source preferred).
  • works on Windows (10).
  • works with a Mono track. I know that some software simply subtracts right and left channel, which often works for commercial music. I don't necessarily have that.
  • works on Udio generated MP3s.

It needn't:

  • deal with multiple singers. Just one voice at a time.

I have tried:

  • UVR GUI. On the song I tried, it correctly extracted the instruments, but it failed with the voice (the file exists, but it contains mostly silence)
  • Audacity's vocal removal and vocal isolation features, but this relies on the stereo channel method I described earlier. If it works for you, find it under Effect / Vocal reduction and isolation... (at the bottom of the effect menu)
Improve this question
4
  • 2
    You can try out github.com/karaokenerds/python-audio-separator (it's a Python CLI project, works for me on Linux and maybe it works flawlessly on other platforms) and AFAIR its GUI front-end: github.com/Anjok07/ultimatevocalremovergui
    – terrorrussia-keeps-killing
    Commented Jul 13 at 15:07
  • @terrorrussia-keeps-killing: I couldn't wait ;-) Unfortunately, UVR was not able to extract the vocals from the music I had. Instruments were fine. Still a great suggestion, it works on some other files.
    – Thomas Weller
    Commented Jul 13 at 18:58
  • 1
    yup, such software is not perfect but it's really great especially comparing to audio editors like Audacity that really rely on channel subtraction. Sometimes audio-separator and UVR extract vocals and instrumentals imperfectly, especially for instruments that may sound like vocals (electronica enriched with samples or synths). I'm not sure if the cause is using the default model, but I haven't ever tried to switch to another one. Perhaps it may change the things.
    – terrorrussia-keeps-killing
    Commented Jul 15 at 8:06
  • 1
    And a note on Audacity. The only thing I could recall it may use is stereo channel subtraction (to be honest, I have never used Vocal Reduction). Maybe you could also try finding multiracks online (usually MOGG) for your music, but you must be very and very lucky (some artists really do that or these might leak). I'm lucky to find a few tracks in multitrack format of my favorite artists (like NIN), and I could enable/disable each instrument and vocals (but some instruments may use the same track, but not my case). Say, you can enable bass, drums, and vocals only, and it sounds interesting. :)
    – terrorrussia-keeps-killing
    Commented Jul 15 at 8:13

1 Answer 1

Reset to default
2

I found vocal-separate to work with Udio generated songs.

However, I could not get the Source Code to work, even I have quite some experience with Python. So I used the pre-compiled release instead.

take a WAV or MP3 file as input.

Both. It works with WAV and MP3, but it will convert MP3 to WAV first (that's an automatic step).

create 2 WAV or MP3 files as output: one for the music, one for the voice.

Depending on the selected AI model, it will generate 2, 4 or 5 files.

works offline (no Internet connection needed).

It is a local application with the AI models on your machine, but it has a Browser-based user interface, accessed via localhost, which is your own computer.

is gratis (Open Source preferred).

It is gratis, even Open Source (GPL 3.0)

works on Windows (10).

Yes, I ran it on Windows 10.

works on Udio generated MP3s.

Yes. I tried that.

works with a Mono track.

Yes. I tried that.

What I don't like:

  • The quality does not seem to be as good as that of UVR GUI. Everything sounds a bit like under water. But it's good enough for me.
  • The web user interface. I think it's a bit inconvenient to use.
Improve this answer

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.