Powerful New Vocal Remover AI - Instructions

[INGSOC]

Sorry for all the confusion guys! I might have to overhaul the whole instruction set in the main thread and do more VM tests. This is usually why I wait to put out new releases until I've worked out the kinks at every angle, which I'm still doing. I only released this v4 model because it's currently my best performing model and I wanted to give everyone a chance to test and play with it until my final releases were complete.

In the meantime, post your errors here and I'll help as much as I can.

Is this still strictly an upgraded instrumental model from multigenre 2? I remember it was mentioned you will be moving on to vocals improvement next. thx!

 

[Anjok]

Is this still strictly an upgraded instrumental model from multigenre 2? I remember it was mentioned you will be moving on to vocals improvement next. thx!

That is still in development. Working out the training parameters has been a challenge and still a work in progress.

 

[NewAgeRipper]

That is still in development. Working out the training parameters has been a challenge and still a work in progress.

You said the new test model to use this: python inference.py --input path/to/an/audio/file. Could you list that command as if there were an actual song in place to convert please? Do we still use the original cd command before that or how? The full instructions would be grand. I really want to see if there will any improvement for removing the vocals for Evanescence - Snow White Queen.

 

[djtayz]

You said the new test model to use this: python inference.py --input path/to/an/audio/file. Could you list that command as if there were an actual song in place to convert please? Do we still use the original cd command before that or how? The full instructions would be grand. I really want to see if there will any improvement for removing the vocals for Evanescence - Snow White Queen.

The command lines haven't changed

python inference.py --"C:\Users\user\Downloads\vocal-remover-v4\danna.wav" --gpu 0 *

*omit if you don't use GPU

 

[INGSOC]

Thank you @Anjok
I'd like to try UB40 Red Red Wine again with the update but I'll just wait for the GUI release.
I'm still trying to figure out which is faster? CPU or GPU on a 2012 Macbook Pro? Takes about 25-30 min for me per song.
CPU-2.5 Intel core i5
GPU-Intel HD graphics 4000

 

[NewAgeRipper]

The command lines haven't changed

python inference.py --"C:\Users\user\Downloads\vocal-remover-v4\danna.wav" --gpu 0 *

*omit if you don't use GPU

This is not working at all for me. I always kept the files in my documents. I always converted flac files. Does it have to be a wav file now? ffmpeg was in stalled a long time ago. I just don't know.

 

[Lensbrank]

Thank you very much again, Anjok!

A friend of mine tested the new model and it turns out that it appears to give better results but also worse results:

For example with Crush 40 - I am... All of Me, the vocals can be heard more compared to the previous model.

With the "His World" Sonic theme, it outputs a really good result with less vocals.

By the way I'm not criticizing anything, just giving what I noticed with it so far (and I know it's still in development)

 

[djtayz]

This is not working at all for me. I always kept the files in my documents. I always converted flac files. Does it have to be a wav file now? ffmpeg was in stalled a long time ago. I just don't know.

Honestly I've always kept my files inside the vocal remover folder, otherwise it doesn't work for me, the file format wouldn't matter, nothing has changed except a different model.

Please make sure inside the Models folder there is only one model, in this case the newest v4.

 

[NewAgeRipper]

Honestly I've always kept my files inside the vocal remover folder, otherwise it doesn't work for me, the file format wouldn't matter, nothing has changed except a different model.

Please make sure inside the Models folder there is only one model, in this case the newest v4.

Lol. I'm sorry. Yes, the vocal remover folder is what always sat in my documents. And yes only one model. I tried the command from every way to Wednesday and all the cmd does is start a new line.

 

[djtayz]

Lol. I'm sorry. Yes, the vocal remover folder is what always sat in my documents. And yes only one model. I tried the command from every way to Wednesday and all the cmd does is start a new line.

Can you try with an older model to see if it's the v4 that's causing issues or something else is wrong?

 

[Anjok]

Thank you very much again, Anjok!

A friend of mine tested the new model and it turns out that it appears to give better results but also worse results:

For example with Crush 40 - I am... All of Me, the vocals can be heard more compared to the previous model.

With the "His World" Sonic theme, it outputs a really good result with less vocals.

By the way I'm not criticizing anything, just giving what I noticed with it so far (and I know it's still in development)

I don't take that as criticism at all! That's actually really good feedback, the kind that helps me improve these models! I've noticed the exact same thing, this model is good at some the other one was poor at and vice versa.

 

[NewAgeRipper]

Can you try with an older model to see if it's the v4 that's causing issues or something else is wrong?

Yeah. I can install the current one listed in the original post to see.

 

[NewAgeRipper]

[MENTION=26971]djtayz[/MENTION] is something going on with Google Colab? The Multi-Genre 2.0 isn't converting. Multi-Genre NP and HP seem to be fine though.

#FIFTH STEP - MULTI-GENRE 2.0

!python multi-genre-secondgen.py --input "/content/drive/Evanescence - Wasted On You.flac" --gpu 0

loading model... done
loading wave source... Traceback (most recent call last):
File "multi-genre-secondgen.py", line 104, in <module>
main()
File "multi-genre-secondgen.py", line 39, in main
args.input, args.sr, False, dtype=np.float32, res_type='kaiser_fast')
File "/usr/local/lib/python3.6/dist-packages/librosa/core/audio.py", line 119, in load
with audioread.audio_open(os.path.realpath(path)) as input_file:
File "/usr/local/lib/python3.6/dist-packages/audioread/__init__.py", line 111, in audio_open
return BackendClass(path)
File "/usr/local/lib/python3.6/dist-packages/audioread/rawread.py", line 62, in __init__
self._fh = open(filename, 'rb')
FileNotFoundError: [Errno 2] No such file or directory: '/content/drive/Evanescence - Wasted On You.flac'

 

[djtayz]

[MENTION=26971]djtayz[/MENTION] is something going on with Google Colab? The Multi-Genre 2.0 isn't converting. Multi-Genre NP and HP seem to be fine though.

#FIFTH STEP - MULTI-GENRE 2.0

!python multi-genre-secondgen.py --input "/content/drive/Evanescence - Wasted On You.flac" --gpu 0

loading model... done
loading wave source... Traceback (most recent call last):
File "multi-genre-secondgen.py", line 104, in <module>
main()
File "multi-genre-secondgen.py", line 39, in main
args.input, args.sr, False, dtype=np.float32, res_type='kaiser_fast')
File "/usr/local/lib/python3.6/dist-packages/librosa/core/audio.py", line 119, in load
with audioread.audio_open(os.path.realpath(path)) as input_file:
File "/usr/local/lib/python3.6/dist-packages/audioread/__init__.py", line 111, in audio_open
return BackendClass(path)
File "/usr/local/lib/python3.6/dist-packages/audioread/rawread.py", line 62, in __init__
self._fh = open(filename, 'rb')
FileNotFoundError: [Errno 2] No such file or directory: '/content/drive/Evanescence - Wasted On You.flac'

As far as I can tell, it's something to do with the audio file, when you upload a song wait a couple of seconds before attempting to convert it, to give Colab time to update. Also try to rename it to a single word if it still doesn't work.

 

[djtayz]

Really odd, I didn't change anything but the new model won't load now

 

[NewAgeRipper]

As far as I can tell, it's something to do with the audio file, when you upload a song wait a couple of seconds before attempting to convert it, to give Colab time to update. Also try to rename it to a single word if it still doesn't work.

Lol, this happened with songs that were already uploaded for weeks. I tried it with several and same result.

 

[NewAgeRipper]

As far as I can tell, it's something to do with the audio file, when you upload a song wait a couple of seconds before attempting to convert it, to give Colab time to update. Also try to rename it to a single word if it still doesn't work.

I just set it all up as new again and it's working. Apparently I caused a corruption in the code string and didn't see it or know what it was. But working fine now.

 

[djtayz]

I just set it all up as new again and it's working. Apparently I caused a corruption in the code string and didn't see it or know what it was. But working fine now.

Glad it's working. There is a way to set up a Colab project without the need of uploading files to your drive, you just open it, run the code, upload your song and then download it, after closing the project the files are deleted. I think it's better and it would be easier to update

 

[NewAgeRipper]

Glad it's working. There is a way to set up a Colab project without the need of uploading files to your drive, you just open it, run the code, upload your song and then download it, after closing the project the files are deleted. I think it's better and it would be easier to update

Yeah? Maybe you can show us.

 

[Anjok]

Hey guys! I'm still working on getting the next round of models released, as well as the GUI changes. I'm trying to iron everything out and make it as easy to use as possible. I'm keeping track of all of the issues everyone has been experiencing.

[MENTION=10167]NewAgeRipper[/MENTION] - Thank you for your raw feedback on the installation process. I've taken all of your concerns into consideration in testing and drafting new documentation for the new version of the A.I.

Regarding the GoogleColab, I've been pretty hands off of it. [MENTION=26971]djtayz[/MENTION] - Although I've confirmed the GUI will never work with GoogleColab, are there any code changes you'd like me to implement that might make the process easier? Let me know and I'll edit the code accordingly.