[GH-ISSUE #843] Finetuning #539

Owner

Originally created by @benc1221 on GitHub (Jul 23, 2023).
Original GitHub issue: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/843

how and where do i finetune my model? my model has a british accent and the tone of the voice is present but not the accent and i was wondering if finetuning the model will help the issue im facing.

Originally created by @benc1221 on GitHub (Jul 23, 2023). Original GitHub issue: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/843 how and where do i finetune my model? my model has a british accent and the tone of the voice is present but not the accent and i was wondering if finetuning the model will help the issue im facing.

bsakaguchi added the

help wanted

question

labels

bsakaguchi commented

Author

Owner

@fumiama commented on GitHub (Jul 24, 2023):

You might mean training. The more steps you train, the more similarity you will get. But you might encounter the overfitting problem depending on your dataset.

@fumiama commented on GitHub (Jul 24, 2023): You might mean `training`. The more steps you train, the more similarity you will get. But you might encounter the overfitting problem depending on your dataset.

bsakaguchi commented

Author

Owner

@benc1221 commented on GitHub (Jul 24, 2023):

You might mean training. The more steps you train, the more similarity you will get. But you might encounter the overfitting problem depending on your dataset.

So I need to add more epoches? I thought the index files hold the accent unless I'm mistaken

@benc1221 commented on GitHub (Jul 24, 2023): > You might mean `training`. The more steps you train, the more similarity you will get. But you might encounter the overfitting problem depending on your dataset. So I need to add more epoches? I thought the index files hold the accent unless I'm mistaken

bsakaguchi commented

Author

Owner

@fumiama commented on GitHub (Jul 24, 2023):

Well, the index file does generate a part of the accent but not the whole. You can try more epochs and a large dataset. Large dataset can increase the coverage of the speaker's accent.

@fumiama commented on GitHub (Jul 24, 2023): Well, the index file does generate a part of the accent but not the whole. You can try more epochs and a large dataset. Large dataset can increase the coverage of the speaker's accent.

bsakaguchi commented

Author

Owner

@benc1221 commented on GitHub (Jul 24, 2023):

if i do 9 hours of data i think it will take a great deal of time to train

@benc1221 commented on GitHub (Jul 24, 2023): if i do 9 hours of data i think it will take a great deal of time to train

bsakaguchi commented

Author

Owner

@fumiama commented on GitHub (Jul 24, 2023):

if i do 9 hours of data i think it will take a great deal of time to train

Sure. I think 30mins to 1 hour is enough but you should keep its quality.

@fumiama commented on GitHub (Jul 24, 2023): > if i do 9 hours of data i think it will take a great deal of time to train Sure. I think 30mins to 1 hour is enough but you should keep its quality.

bsakaguchi commented

Author

Owner

@benc1221 commented on GitHub (Jul 24, 2023):

i had done 1 hour and it still had not retain its accent

@benc1221 commented on GitHub (Jul 24, 2023): i had done 1 hour and it still had not retain its accent

bsakaguchi commented

Author

Owner

@benc1221 commented on GitHub (Jul 25, 2023):

ok so theres D_2333333.pth and G_2333333.pth
what is the difference between the two? i know they mean discriminative and generative but and D_2333333.pth is has a bigger file size than G_2333333.pth

@benc1221 commented on GitHub (Jul 25, 2023): ok so theres D_2333333.pth and G_2333333.pth what is the difference between the two? i know they mean discriminative and generative but and D_2333333.pth is has a bigger file size than G_2333333.pth

bsakaguchi commented

Author

Owner

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

the final weight will be extracted from G. G and D are used for continuing training.

@RVC-Boss commented on GitHub (Jul 27, 2023): the final weight will be extracted from G. G and D are used for continuing training.

bsakaguchi commented

Author

Owner

@benc1221 commented on GitHub (Aug 6, 2023):

the final weight will be extracted from G. G and D are used for continuing training.

the problem im having is having a high quality dataset and when i use index file it uses some of the accent but not strongly at all its just mainly the input speakers native accent. i trained for 1000 epochs even 100000 epochs and still having the same issue. i just need help with retaining the accent for each voice i make.

@benc1221 commented on GitHub (Aug 6, 2023): > the final weight will be extracted from G. G and D are used for continuing training. the problem im having is having a high quality dataset and when i use index file it uses some of the accent but not strongly at all its just mainly the input speakers native accent. i trained for 1000 epochs even 100000 epochs and still having the same issue. i just need help with retaining the accent for each voice i make.

bsakaguchi commented

Author

Owner

@sakiodre commented on GitHub (Oct 22, 2023):

Same question, i want to change the accent of the input audio to my model, can't figure it out

@sakiodre commented on GitHub (Oct 22, 2023): Same question, i want to change the accent of the input audio to my model, can't figure it out

bsakaguchi commented