search

flashlight / wav2letter

Facebook AI Research's Automatic Speech Recognition Toolkit
https://github.com/facebookresearch/wav2letter/wiki
Other
6.39k stars 1.01k forks source link

Suggestion on modifying the network architecture #86

Closed sinhnn closed 6 years ago

sinhnn commented 6 years ago

Dear Sir, I am planning using the wav2letter for vietnamese. I have read the network architecture arch/librispeech-glu-highdropout, it has many layers. I would like to change the architecture. Could you give me any suggestions? Thank you so much.

vineelpratap commented 6 years ago

Hi, You can use the same building blocks as in the librispeech architecture (Conv layers followed by GLUs). If your dataset is small, try to reduce the width of each layer and the total number of layers to avoid overfitting.

sinhnn commented 6 years ago

Dear Sir, Thank you, I will try as your comments.