sru/speech at master · asappresearch/sru · GitHub

Name		Name	Last commit message	Last commit date
parent directory ..
BLSTM_5L.ndl		BLSTM_5L.ndl
Bi-SRU_12L.ndl		Bi-SRU_12L.ndl
CNTK_Bi-SRU.config		CNTK_Bi-SRU.config
CNTK_Bi-SRU_smbr.config		CNTK_Bi-SRU_smbr.config
CNTK_SRU.config		CNTK_SRU.config
CNTK_SRU_smbr.config		CNTK_SRU_smbr.config
LSTM_5L.ndl		LSTM_5L.ndl
README.md		README.md
SRU_12L.ndl		SRU_12L.ndl
make_fbank_pitch.sh		make_fbank_pitch.sh
run_bi-sru.sh		run_bi-sru.sh
run_sru.sh		run_sru.sh
run_swbd.sh		run_swbd.sh

README.md

Code used for speech recognition tasks. We evaluate LSTM and SRU on SWBD task.

How to run

Build Kaldi from Kaldi.
Build CNTK from yzhang87/cntk.
Build KaldiReader in CNTK (follow the instruction).
Build SWBD baseline (Set proper path in run.sh):

  run.sh

- Run run_SRU.sh and get the results.

  run_SRU.sh --ndl ndlfile

TODO

Make it compatible with newest CNTK version.
Write custimized kernel for SRU.
Port to Pytorch and MXNet.