M1 Mac : Unable To Run | Version : 1.5.3.9

[Ghostkwebb]

Issue Type

Bug Report

vc client version number

MMVCServerSIO_mac_onnxcpu-nocuda_v.1.5.3.10.zip

OS

MacOS Sonoma

GPU

M1 Pro

Clear setting

no

Sample model

no

Input chunk num

no

Wait for a while

No

read tutorial

yes

Extract files to a new folder.

no

Voice Changer type

RVC

Model type

Not sure

Situation

I tried everything from the tutorial but this keeps happening. Do you know what I can do?

https://github.com/w-okada/voice-changer/assets/61558762/8aa394db-8b53-4831-808a-5170c9e28a8f

[Ghostkwebb]

c58657da-d123-418f-8baa-7b796ac44b8c.mp4

Also what about this?

[atrutlebtw]

yeah im having similar audio issues when running v3

[Ghostkwebb]

Ready for experiments!

[w-okada]

I'm very grateful. However, I only have one idea right now. Please try using this .command. startHttp4.zip

[ChinHongTan]

I am able to run with v3, and by increasing chunk size (1024) I do not have any audio issues

[w-okada]

Please try v4.

[atrutlebtw]

will do

[Ghostkwebb]

Gimme a second I will give it a try

[atrutlebtw]

just white noise, gets slightly louder with microphone input, tried chunk sizes 512 - 2048 and dio, crepe and harvest, all with the same result also tried using a audio file instead of microphone, same result

[ChinHongTan]

Can confirm v4 gets a big performance improvement over v3. With v3 I get res time between 1500-2000ms while v4 is around 400-500ms.

[atrutlebtw]

yeah agreed, just no real voicing for me

[Ghostkwebb]

just white noise, gets slightly louder with microphone input, tried chunk sizes 512 - 2048 and dio, crepe and harvest, all with the same result also tried using a audio file instead of microphone, same result

It's working fine for you??

[Ghostkwebb]

Can confirm v4 gets a big performance improvement over v3. With v3 I get res time between 1500-2000ms while v4 is around 400-500ms.

I still get that wierd voice when trying

[Ghostkwebb]

Is it really just me?

https://github.com/w-okada/voice-changer/assets/61558762/a828824b-350d-4829-af26-c3904b12d6fd

[w-okada]

@Ghostkwebb 1 question. Have you installed pytorch by yourself? (for the other application.) Perhaps, the previous pytorch may cause the result.　It's just one possibility.

[Ghostkwebb]

@Ghostkwebb 1 question. Have you installed pytorch by yourself? (for the other application.) Perhaps, the previous pytorch may cause the result.

Yes I did install pytorch by myself for other applications

[Ghostkwebb]

[w-okada]

I see. However, it is a problem whether or not the previous version of PyTorch effects.

[Ghostkwebb]

What should I do then?

[w-okada]

can you exec

ls /usr/local/lib/libtorchaudio.so
ls /usr/lib/libtorchaudio.so

[Ghostkwebb]

Doesn't exist

[Ghostkwebb]

[w-okada]

really? so, previous pytorch is not cause... sorry.

[Toby1009]

Is it really just me?真的只有我一個人這樣嗎？

Screen.Recording.2023-07-25.at.6.58.41.PM.mp4

It's fine for me , no weird noise

[Ghostkwebb]

WHAT

[Ghostkwebb]

WHY ME GOD WHY ME!

[w-okada]

I am generating experiment package with v4. Let's check if there are others experiencing the same phenomenon.

[w-okada]

The users in Simillar situation is here https://github.com/w-okada/voice-changer/issues/339

[Ghostkwebb]

The users in Simillar situation is here #339

Everybody there is either on Monterey or Ventura but I'm on MacOS Sonoma so my bad it may be because my os is. experimental so sorry about that the noise thing is most probably a Sonoma thing which is similar to obs that I already reported. Thank you w-okada for all the support towards people on mac u can close this whenever u want now!

[kelvin-zhao]

v3 and v4 seems to work on Monterey, but I tried uploading a trained voice, the results doesn't seem to sound anything like it and it sounds pretty weird. So does the default voices, all sound super weird.

[w-okada]

@kelvin-zhao how about sample model?

[kelvin-zhao]

Yea, sample sounds really weird too, the pitch goes up and down, sounds really weird.

[w-okada]

really? I have a feeling that torchaudio isn't working very well, so I guess I'll have to consider not using it...

[w-okada]

I tried creating a version without using torchaudio as an experiment. Could you give it a try?

https://huggingface.co/wok000/vcclient000/resolve/main/MMVCServerSIO_mac_onnxcpu-nocuda_v.1.5.3.11_alpha.zip

[UncleDreiAI]

i was able to get 1.5.3.9 to run on a M1 but it's like a delay and/or glitchy...with echoes/delay effect does that mean my macbook is tapping out :-(

using the starthttp_v4 command

trying 1.5.3.11 gives me error: "MMVCServerSIO_mac_onnxcpu-nocuda_v.1.5.3.11_alpha/startHttp.command: line 35: 67014 Killed: 9 "

[w-okada]

@dredot Sorry, instead of using startHttp.command, please use startHttpEnv.command.

[UncleDreiAI]

'Sorry, instead of using startHttp.command, please use startHttpEnv.command.'

This is what i'm seeing. :-(

[w-okada]

Did it not work out? I don't understand.

Could you please try using this .command file? If this doesn't work, I really won't know what to do next. startHttp5.zip

I don't understand why I can't reproduce the bug in my own environment for the most part.

[UncleDreiAI]

Did it not work out? I don't understand.

Could you please try using this .command file? If this doesn't work, I really won't know what to do next. startHttp5.zip

I don't understand why I can't reproduce the bug in my own environment for the most part.

for 11-alpha i am still getting the error from above... i will go back to .9 and then i we can figure out if my computer is even capable to give us real time

[UncleDreiAI]

Did it not work out? I don't understand.

Could you please try using this .command file? If this doesn't work, I really won't know what to do next. startHttp5.zip

I don't understand why I can't reproduce the bug in my own environment for the most part.

1.5.3.9 works just fine with your startHTTP update... the only problem i'm having now is a delay and then sometimes it goes in to a glitchy echo loop

:-(

[Ghostkwebb]

Did it not work out? I don't understand.

Could you please try using this .command file? If this doesn't work, I really won't know what to do next. startHttp5.zip

I don't understand why I can't reproduce the bug in my own environment for the most part.

Which MacOS are you on @w-okada ?

[w-okada]

[Ghostkwebb]

Well u should upgrade to ventura/sonoma cause I think most people are on it (at least the people here) and that's why you are unable to recreate the problem (I think at least)

[ramsemaj]

when I ran startHttp5, it exited with 137 which is out of memory (according to the internet). i get the same when trying to run manually.

% sh startHttp5.command
[Voice Changer] matplotlib is dummied
startHttp5.command: line 19:  4246 Killed: 9               PYTORCH_ENABLE_MPS_FALLBACK=1 MMVCServerSIO.app/Contents/MacOS/MMVCServerSIO --https off -p 18888 --content_vec_500 pretrain/checkpoint_best_legacy_500.pt --content_vec_500_onnx pretrain/content_vec_500.onnx --content_vec_500_onnx_on true --hubert_base pretrain/hubert_base.pt --hubert_base_jp pretrain/rinna_hubert_base_jp.pt --hubert_soft pretrain/hubert-soft-0d54a1f4.pt --nsf_hifigan pretrain/nsf_hifigan/model --crepe_onnx_full pretrain/crepe_onnx_full.onnx --crepe_onnx_tiny pretrain/crepe_onnx_tiny.onnx --rmvpe pretrain/rmvpe.pt --model_dir model_dir --samples samples.json

% echo $?                                        
137

[NBNew]

I tried creating a version without using torchaudio as an experiment. Could you give it a try?

https://huggingface.co/wok000/vcclient000/resolve/main/MMVCServerSIO_mac_onnxcpu-nocuda_v.1.5.3.11_alpha.zip

Hey @w-okada , I would like to inform you that it indeed worked with with demo ( without torchaudio ), however the voice is soo choppy that it's difficult to understand even a single word and I guess thats expected since torchaudio is missing ? For reference, I am on attached version

[Ghostkwebb]

I tried creating a version without using torchaudio as an experiment. Could you give it a try? https://huggingface.co/wok000/vcclient000/resolve/main/MMVCServerSIO_mac_onnxcpu-nocuda_v.1.5.3.11_alpha.zip

Hey @w-okada , I would like to inform you that it indeed worked with with demo ( without torchaudio ), however the voice is soo choppy that it's difficult to understand even a single word and I guess thats expected since torchaudio is missing ? For reference, I am on attached version

Does it sound like this? https://user-images.githubusercontent.com/61558762/255915277-a828824b-350d-4829-af26-c3904b12d6fd.mp4

[NBNew]

I tried creating a version without using torchaudio as an experiment. Could you give it a try? https://huggingface.co/wok000/vcclient000/resolve/main/MMVCServerSIO_mac_onnxcpu-nocuda_v.1.5.3.11_alpha.zip

Hey @w-okada , I would like to inform you that it indeed worked with with demo ( without torchaudio ), however the voice is soo choppy that it's difficult to understand even a single word and I guess thats expected since torchaudio is missing ? For reference, I am on attached version

Does it sound like this? https://user-images.githubusercontent.com/61558762/255915277-a828824b-350d-4829-af26-c3904b12d6fd.mp4

That was sooo funny :D , it's like someone is humming. No, for me I have the wordings but its not clear enough, not tuned. May be something to do with settings. Could you recommend some settings to try out ?

[Ghostkwebb]

Damn it really is a macOS sonoma problem ooof

[NBNew]

Damn it really is a macOS sonoma problem ooof

but the other versions didn't work for me, got dummied related error while starting it. Please recommend good settings if possible, I would like to see if its perfectly working with v11 or not

[Ghostkwebb]

Other versions work for me when the startupHttp 3/4 is used but the humming sound comes so idk what the problem is. The last usable version I used was 1.5.3.5a and that worked perfectly.