Automatic Speech Recognition (ASR) system trained on CommonVoice (zh-TW) dataset with Kaldi toolkit.

Simply run shrun.sh to train and test the three models below:

1. Monophone
2. Triphone (1st pass): Delta + Delta-Delta
3. Triphone (2nd pass): LDA + MLLT

scripts/prepare_data.py : Preprocess CommonVoice (zh-TW) for usage of Kaldi.
scripts/prepare_data.ipynb : Full details and explanations of the prepare_data.py.

See requirements.txt to check if any required package is not installed.