kaldi_taiwanese

Deep Learning taiwanese recognization homework 1 at NTUT.

Step1. go to kaldi workspace

$ cd ~/kaldi/egs/kakdi_taiwanese/s5

Step2. create bash file with convert audo sample rate .

$ gedit  wav22k_16k.sh

$ chmod +x wav22k_16k.sh

$ ./wav22k_16k.sh

result: generate 16k audo wave file to train/wav and test/wav.

step3. create python file: train.csv convert to train.txt

$ chmod +x HW1_csv2text.py

$ pyhton HW1_csv2text.py

step4. copy to lexicon dir after lexicon.txt delete last raw, copy to data/train and data/test dir after train.txt rename is text .

step5. edit run_tdnn.sh, set network and gpu num.

$ gedit local/chain/run_tdnn.sh

step6. run run.sh

$ chmod +x run.sh

$ ./run.sh