search

r9y9 / tacotron_pytorch

PyTorch implementation of Tacotron speech synthesis model.
http://nbviewer.jupyter.org/github/r9y9/tacotron_pytorch/blob/master/notebooks/Test%20Tacotron.ipynb
Other
306 stars 79 forks source link

why turning off dropout of decoder's prenet make a serious performance regression #20

Closed zwlanpishu closed 5 years ago

zwlanpishu commented 5 years ago

Thanks very much for your codes. When i was working with it, also found that dropout on eval make a serious regresssion. I can not find out where is the problem. May it caused by the combination of BN and dropout?

pravn commented 5 years ago

Afaik, Dropout helps in generalizing to unseen examples. I experience a lot in transfer learning examples (esp when data is scarce) in that when the prenet (basically full connections plus dropout) performance is abysmal.

pravn.wordpress.com

On Wed, Jul 3, 2019, 7:00 PM zwlanpishu notifications@github.com wrote:

Thanks very much for your codes. When i was working with it, also found that dropout on eval make a serious regresssion. I can not find out where is the problem. May it caused by the combination of BN and dropout?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/r9y9/tacotron_pytorch/issues/20?email_source=notifications&email_token=AAEGOVLFX4A2SCJWL5Q4NZDP5VKSDA5CNFSM4H5PZWSKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4G5IOCGQ, or mute the thread https://github.com/notifications/unsubscribe-auth/AAEGOVJFBD2FGRHV3NBRYKLP5VKSDANCNFSM4H5PZWSA .

stale[bot] commented 5 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.