chiayuli/pdnnlg_asr
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
### Tools Must be installed ### install_pfile_utils.sh ### 300 hours Switchboard Task ### swbd folder includes the recipes of training DNN/CNN/CNN-LACE based Acoustic Models with SWBD Data Set for ASR system. In contains run_swbd_pdnnlg/ pdnnlg/ steps_pdnnlg/ Please copy all three folders under Kaldi-trunck/egs/swbd/s5c/ , and follow these steps: 1) Download/Install Kaldi 2) Use $KALDI_ROOT/egs/swbd/s5c/run.sh recipe to get GMM-HMMs SAT model (tri4) 3) Use run_swbd/pdnnlg/run-dnn.sh recipe to train DNN-HMMs model with fMLLR features, and do decoding run_swbd/pdnnlg/run-dnn-fbank.sh recipe to train DNN-HMMs model with filter-bank features, and do decoding run_swbd/pdnnlg/run-cnn.sh recipe to train CNN-HMMs model with filter-bank features, and do decoding run_swbd/pdnnlg/run-cnn-lacea.sh recipe to train CNN-LACE-HMMs model with filter-bank, and do decoding ### 80 hours Wall Street Journal Task ### wsj folder includes the recipes of training DNN/CNN/CNN-LACE based Acoustic Models with WSJ Data Set for ASR system In contains run_wsj_pdnnlg/ pdnnlg/ steps_pdnnlg/ Please copy all three folders under Kaldi-trunck/egs/wsj/s5/ , and follow these steps: 1) Download/Install Kaldi 2) Use $KALDI_ROOT/egs/wsj/s5/run.sh recipe to get GMM-HMMs SAT model (tri4b) 3) Use run_wsj/pdnnlg/run-dnn.sh recipe to train DNN-HMMs model with fMLLR features, and do decoding run_wsj/pdnnlg/run-dnn-fbank.sh recipe to train DNN-HMMs model with filter-bank featuers, and do decoding run_wsj/pdnnlg/run-cnn.sh recipe to train CNN-HMMs model with filter-bank, and do decoding run_wsj/pdnnlg/run-cnn-lacea.sh recipe to train CNN-LACE-HMMs model with filter-bank, and do decoding run_wsj/pdnnlg/run-densenet.sh recipe to train DENSENET-HMMs model with filter-bank, and do decoding
About
The implementation of DNN, CNN, LACE and DenseNet for acoustic models in ASR systems. The code is written in Python, Theano and Lasagne.
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published