pytorch-kaldi is a project for developing state-of-the-art DNN/RNN hybrid speech recognition systems. The DNN part is managed by pytorch, while feature extraction, label computation, and decoding are performed with the kaldi toolkit.
I have tried to perform testing from a trained network but failed to have WER output in some cases. I think there might be some wrong operation in decoding.
I have compared it with the successful cases and found out this file 'exp/20201021_Pattern_Search/TIMIT_LSTM_fmllr_L2_8bw_16ba_16_8x8x_4/decode_TIMIT_test_out_dnn2/score_1/ctm_39phn.filt.sys' doesn't exist in failed case. Can you help me with which step(code) will generate this file?
#file in successful exp/20201021_Pattern_Search/TIMIT_LSTM_fmllr_L2_8bw_16ba_16_8x8x_4/decode_TIMIT_test_out_dnn2/score_1:
![image](https://user-images.githubusercontent.com/33325753/97245444-04a2a280-1836-11eb-9c92-d9f6362e71e1.png)
#file in failed exp/20201021_Pattern_Search/TIMIT_LSTM_fmllr_L2_8bw_16ba_16_8x8x_8/decode_TIMIT_test_out_dnn2/score_1:
![image](https://user-images.githubusercontent.com/33325753/97245499-1f751700-1836-11eb-9c56-57b4c1312d9c.png)
I have tried to perform testing from a trained network but failed to have WER output in some cases. I think there might be some wrong operation in decoding. I have compared it with the successful cases and found out this file 'exp/20201021_Pattern_Search/TIMIT_LSTM_fmllr_L2_8bw_16ba_16_8x8x_4/decode_TIMIT_test_out_dnn2/score_1/ctm_39phn.filt.sys' doesn't exist in failed case. Can you help me with which step(code) will generate this file?