search

tsurumeso / vocal-remover

Vocal Remover using Deep Neural Networks
MIT License
1.47k stars 215 forks source link

Wanted to Let Everyone Know a GUI Was Created for This Vocal Remover! #38

Open Anjok07 opened 3 years ago

Anjok07 commented 3 years ago

This isn't an issue, but I wanted to let you and the community here know. This GUI is 100% based on your vocal remover, options included. This was a joint project between another coder and myself. Feel free to use, edit, and implement as you wish! Great job on this AI!

I also included 2 additional models that I trained myself. One trained on 700 pairs and another trained on trained data. Both of them came out GREAT!

You can access it via the following:

Zcooger commented 3 years ago

The Multi-Genre model is sick, cannot wait for next updates. Keep the training on! No tool made such good quality separation yet like this. Props to both of you! I noticed that organ (mostly hammond), some electric guitars, trumpets, electronic sythesizers and saxophones are treated as vocals while backing voices are leaking to instruments but it's editable.

Anjok07 commented 3 years ago

The Multi-Genre model is sick, cannot wait for next updates. Keep the training on! No tool made such good quality separation yet like this. Props to both of you! I noticed that organ (mostly hammond), some electric guitars, trumpets, electronic sythesizers and saxophones are treated as vocals while backing voices are leaking to instruments but it's editable.

Thank you! I released new models that don't bleed like the one that comes with it does. Check it out!

I'm also in the process of training new models that are set to outperform all of the ones I've released thus far.

Zcooger commented 3 years ago

To be honest didn't notice the extra package, gonna watch out. Got an idea to unlearn the model to treat some of the instruments as voice: get an track that has no any traces of words/singing and put it in instrumentals and mixtures. Good examples of bands with difficult instruments are 4Bars, Bert Weedon, Hillary Thaddeus, Jack McDuff, Mezzoforte, T-Square, Walter Wanderley... Vice versa for voice/podcast/narration recordings where instrumental contains silent track and mixture the voice only. Some songs contain regular speaking and melodeclamation. Good resource for free songs in lossless FLAC format (and MP3 too): https://free-mp3-download.net/ . I have a MUSDB18-HQ multitrack music data set if you want it to be sent to you (it weighs 22GB).

Anjok07 commented 3 years ago

To be honest didn't notice the extra package, gonna watch out. Got an idea to unlearn the model to treat some of the instruments as voice: get an track that has no any traces of words/singing and put it in instrumentals and mixtures. Good examples of bands with difficult instruments are 4Bars, Bert Weedon, Hillary Thaddeus, Jack McDuff, Mezzoforte, T-Square, Walter Wanderley... Vice versa for voice/podcast/narration recordings where instrumental contains silent track and mixture the voice only. Some songs contain regular speaking and melodeclamation. Good resource for free songs in lossless FLAC format (and MP3 too): https://free-mp3-download.net/ . I have a MUSDB18-HQ multitrack music data set if you want it to be sent to you (it weighs 22GB).

That would be awesome! I would love to try that dataset

Zcooger commented 3 years ago

https://mega.nz/file/OVgRXZaS#KzNQ6zvd7qMmd9mraOj8Ow29N3v7EAMjFC6WQGjVro0

Anjok07 commented 3 years ago

https://mega.nz/file/OVgRXZaS#KzNQ6zvd7qMmd9mraOj8Ow29N3v7EAMjFC6WQGjVro0

Thanks!

aufr33 commented 3 years ago

@Zcooger Thanks.

tcafranz commented 3 years ago

I can't get this thing to work... having trouble installing the prereq's. Before I waste time typing my issue, does anyone care to help me out?

Anjok07 commented 3 years ago

You'll need to provide details. What's the error and what OS are you using?

Zcooger commented 3 years ago

While working with image datasets I got an idea to increase dataset variety by reversing the audio samples in time.