search

NVIDIA / OpenSeq2Seq

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP
https://nvidia.github.io/OpenSeq2Seq
Apache License 2.0
1.54k stars 372 forks source link

Config file for Fisher+Switchboard? #488

Closed Shujian2015 closed 4 years ago

Shujian2015 commented 4 years ago

Hi, thanks for open source this awesome package. Is there any plan to add the config file for Fisher+Switchboard datasets?

borisgin commented 4 years ago

Currently we don't plan to add config file for Fisher-SWB. But PR for downloading and training CommonVoice would be welcomed :)

Shujian2015 commented 4 years ago

Hi @borisgin, thanks for your response. Is there any particular reason that for fisher_swb data, Jasper DR 10x5 was trained using SGD with momentum but not NovoGrad and it is only trained for 50 epochs instead of 400?

borisgin commented 4 years ago

No particular reason, just lack of time before Interspeech deadline

Shujian2015 commented 4 years ago

Thanks again @borisgin !

Shujian2015 commented 4 years ago

Hi @borisgin, how did you handle things like , [breath], [cough] in this dataset? Did you put them into vocab, or just delete them, or any other methods? Thanks in advance.