Changing voice causes tune change (server mode)

[The-Hipnotist]

Issue Type

Bug Report

vc client version number

MMVCServerSIO_win_onnxdirectML-cuda_v.1.5.3.12a

OS

Windows 11

GPU

NVIDIA RTX 3050

Clear setting

yes

Sample model

yes

Input chunk num

yes

Wait for a while

The GUI successfully launched.

read tutorial

yes

Extract files to a new folder.

yes

Voice Changer type

RVC

Model type

It depends on the model, but for the most part its pytorch.

Situation

Whenever I switch to a new voice (in server mode and without changing the chunk size), the tune changes suddenly to a deeper tune than the one I set originally. This was supposedly fixed in the newer version (which I have), but the issue still persists.

If I restart the program with the untuned voice however, it fixes itself and goes back to the original tune, but once I switch to another voice, it persists.

I am currently trying the onnxgpu version as opposed to the onnxdirectML version (since i have an nvidia gpu). I'll close and update this issue if it fixed it.

application window capture

No response

logs on terminal

SlotInfo::: RVCModelSlot(slotIndex=9, voiceChangerType='RVC', name='GDColon', description='', credit='', termsOfUseUrl='', iconFile='model_dir\9\1lV9ZaGS_400x400.jpg', speakers={'0': 'target'}, modelFile='GDColon.pth', indexFile='added_IVF2901_Flat_nprobe_1_GDColon_v2.index', defaultTune=-2, defaultIndexRatio=0, defaultProtect=0.5, isONNX=False, modelType='pyTorchRVCv2', samplingRate=48000, f0=True, embChannels=768, embOutputLayer=12, useFinalProj=False, deprecated=False, embedder='hubert_base', sampleId='')

This is an output of my saved settings for one of my models. It should stay at -2 tune, but it sounds more like -5 or -10.

[w-okada]

I can not reproduce.

[The-Hipnotist]

Just tried the onnxcuda version, that one work's without the tune change. I must've gotten the wrong version.

[The-Hipnotist]

I stand corrected, it is doing it again with the same voice, once i switch to it, all the other voices are off tune.

Edit: If I dont switch to the affected voice, every other voice is fine.

[w-okada]

maybe your model is nof0 model. check it.

[The-Hipnotist]

how do i do that?

[w-okada]

check here

[The-Hipnotist]

This is what it says for the model:

[The-Hipnotist]

Edit: Just tried to reimport the models, its still doing the same thing

[w-okada]

no time to reproduce...