0. Environment

0-1. 已安裝Espnet & KALDI

cd s5-esp
nano path.sh
# 把$MAIN_ROOT修改成espnet的路徑
ln -s $ESPNET_ROOT/tools/kaldi/egs/wsj/s5/steps
ln -s $ESPNET_ROOT/tools/kaldi/egs/wsj/s5/utils
cd -

cd s5-kaldi
nano path.sh
# 把$KLADI_ROOT修改成kaldi的路徑
ln -s $KALDI_ROOT/egs/wsj/s5/steps
ln -s $KALDI_ROOT/egs/wsj/s5/utils
cd -

1. Data preparation

1-1. data directory

data/spoken_test_2022_jan28

1-2. espnet

cd s5-esp
local/prep/prepare_data.sh --stage -2 --stop-stage -1

1-2. kaldi (alternative)

cd s5-kaldi
local/prep/prepare_data_kaldi.sh --stage -2 --stop-stage -1

2. Feature extraction

2-1. espnet

cd s5-esp
local/prep/prepare_data.sh --data_name data/spoken_test_2022_jan28

2-1. kaldi (alternative)

cd s5-kaldi
local/prep/prepare_data_kaldi.sh --data_name data/spoken_test_2022_jan28

3. Grader

cd grader
python local/stats_models/multivar_linear_regression.py

4. Notes

修改成自己conda的啟動方式:

可以1)直接刪掉，在執行script前先啟動或2)替換成自己的conda並搭配conda activte $env_name

https://github.com/teinhonglo/dhe_spoken_test/blob/main/s5-esp/local/prep/prepare_data.sh#L28
https://github.com/teinhonglo/dhe_spoken_test/blob/main/s5-esp/local/prep/prepare_data_kaldi.sh#L29
https://github.com/teinhonglo/dhe_spoken_test/blob/main/s5-esp/local/kaldi_stt/extract_feats.sh#L150