[GH-ISSUE #867] model fusion sounds robotic\pitchy? #557

Closed
opened 2026-10-20 10:00:36 +00:00 by bsakaguchi · 5 comments
Owner

Originally created by @Yogevsho on GitHub (Jul 25, 2023).
Original GitHub issue: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/867

hey, I followed the instructions on combining 2 voice models, and no matter what 2 types of models im combining - the new model always comes out robotic and un-usable.
i pointed the gui to right files (both .pth files of the models from the 'weight' folders), and theyre both are V1.
am i doing something wrong here?
also, the merge doesn't produce a new 'added' file, so i tried (both of) the 'added' files of the original models that ive combined - maybe this is the problem?

thank you in advance for your help!

Originally created by @Yogevsho on GitHub (Jul 25, 2023). Original GitHub issue: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/867 hey, I followed the instructions on combining 2 voice models, and no matter what 2 types of models im combining - the new model always comes out robotic and un-usable. i pointed the gui to right files (both .pth files of the models from the 'weight' folders), and theyre both are V1. am i doing something wrong here? also, the merge doesn't produce a new 'added' file, so i tried (both of) the 'added' files of the original models that ive combined - maybe this is the problem? thank you in advance for your help!
bsakaguchi 2026-10-20 10:00:36 +00:00
Author
Owner

@RVC-Boss commented on GitHub (Jul 26, 2023):

The "if_f0"and sample_rate config should also be the same

<!-- gh-comment-id:1650908979 --> @RVC-Boss commented on GitHub (Jul 26, 2023): The "if_f0"and sample_rate config should also be the same
Author
Owner

@Yogevsho commented on GitHub (Jul 26, 2023):

The "if_f0"and sample_rate config should also be the same

You mean the sample_rate config when im training the original model?
and as for the if_f0 - do you mean the ‘whether the model have pitch guidance’ (Also in each model training part)?

thank you!

<!-- gh-comment-id:1651877093 --> @Yogevsho commented on GitHub (Jul 26, 2023): > The "if_f0"and sample_rate config should also be the same You mean the sample_rate config when im training the original model? and as for the if_f0 - do you mean the ‘whether the model have pitch guidance’ (Also in each model training part)? thank you!
Author
Owner

@RVC-Boss commented on GitHub (Jul 27, 2023):

The "if_f0"and sample_rate config should also be the same

You mean the sample_rate config when im training the original model? and as for the if_f0 - do you mean the ‘whether the model have pitch guidance’ (Also in each model training part)?

thank you!

yes

<!-- gh-comment-id:1652859334 --> @RVC-Boss commented on GitHub (Jul 27, 2023): > > The "if_f0"and sample_rate config should also be the same > > You mean the sample_rate config when im training the original model? and as for the if_f0 - do you mean the ‘whether the model have pitch guidance’ (Also in each model training part)? > > thank you! yes
Author
Owner

@Yogevsho commented on GitHub (Jul 28, 2023):

The "if_f0"and sample_rate config should also be the same

You mean the sample_rate config when im training the original model? and as for the if_f0 - do you mean the ‘whether the model have pitch guidance’ (Also in each model training part)?
thank you!

yes

Well i did everything you said and it still comes out robotic. I also made sure and checked the models for matching settings beforehand. Still no good.
Just for clearance - i chose the .pth files of the models from the weights folders (the small ones), and not those from the ‘logs’ folder. Is that ok? or that for the fusion i need both of the models big .pth files (‘G333…’ or ‘D333…’)?

<!-- gh-comment-id:1655766883 --> @Yogevsho commented on GitHub (Jul 28, 2023): > > > The "if_f0"and sample_rate config should also be the same > > > > > > You mean the sample_rate config when im training the original model? and as for the if_f0 - do you mean the ‘whether the model have pitch guidance’ (Also in each model training part)? > > thank you! > > yes Well i did everything you said and it still comes out robotic. I also made sure and checked the models for matching settings beforehand. Still no good. Just for clearance - i chose the .pth files of the models from the weights folders (the small ones), and not those from the ‘logs’ folder. Is that ok? or that for the fusion i need both of the models big .pth files (‘G333…’ or ‘D333…’)?
Author
Owner

@ms903x1 commented on GitHub (Jul 31, 2023):

The model fusion function only supports PTH under the weights folder. If the inference performance of the fused model is poor when all configurations are correct, the problem may be that the two models to be fused are bad.

<!-- gh-comment-id:1657525399 --> @ms903x1 commented on GitHub (Jul 31, 2023): The model fusion function only supports PTH under the weights folder. If the inference performance of the fused model is poor when all configurations are correct, the problem may be that the two models to be fused are bad.
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
RVC-Project/Retrieval-based-Voice-Conversion-WebUI#557
No description provided.