Skip to content

end-to-end automatic speech recognition model with ctc loss

Notifications You must be signed in to change notification settings

edchengg/End-to-End_Model_CTC

Repository files navigation

End-to-End Automatic Speech Recognition(ASR) CTC

This repository contains baseline models(3-5 layers Bi-LSTM) for ASR tasks on standard speech datasets(TIMIT, WSJ, Switchboard).

Model

3 layers or 5 layers BiLSTM + Softmax Layer + CTC Loss

Dataloader

3 dataloaders for 3 different datasets

Results

Switchboard WSJ TIMIT
Dev 11.86(CER) 6.1(CER) 13.429(PER)
Test 4.6(CER) 15.967(PER)

Visualization

Visualization of LSTM hidden units before pretraining and after pretraining.

Alt Text Alt Text

About

end-to-end automatic speech recognition model with ctc loss

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published