search

onejiin / CycleGAN-VC2

CycleGAN-VC2: Improved CycleGAN-based Non-parallel Voice Conversion
MIT License
41 stars 7 forks source link

Is epoch = 5000 too large? #3

Open lightwithshadow opened 5 years ago

lightwithshadow commented 5 years ago

Thank you for your contribution to this project. When I was running demo, an epoch take about 30 minutes. Is epoch = 5000 too large? Looking forward to your reply! Thanks!!!

onejiin commented 5 years ago

Hi,

I think 5000epoch are quite large, thus overfitting may be occurring. Code includes tensorboard as "--tensorboard_log_dir". I don’t know the database what you use, but when validation set inferences, learning rate and training epoch can be adjusted by checking the MSE curve. For VCC2016 database, I will update the learning curve.

Cheers, Wonjin

lightwithshadow commented 5 years ago

Hi,

Thanks for your reply! I will try it again. Thanks for your attention to this matter.

Cheers, shadow

------------------ 原始邮件 ------------------ 发件人: "Wonjin, Jung"<notifications@github.com>; 发送时间: 2019年9月23日(星期一) 中午1:25 收件人: "onejiin/CycleGAN-VC2"<CycleGAN-VC2@noreply.github.com>; 抄送: "1101174181"<1101174181@qq.com>;"Author"<author@noreply.github.com>; 主题: Re: [onejiin/CycleGAN-VC2] Is epoch = 5000 too large? (#3)

Hi,

I think 5000epoch are quite large, thus overfitting may be occurring. Code includes tensorboard as "--tensorboard_log_dir". I don’t know the database what you use, but when validation set inferences, learning rate and training epoch can be adjusted by checking the MSE curve. For VCC2016 database, I will update the learning curve.

Cheers, Wonjin

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.