Created by: pkuyym

1. Use codecs to make sure that all texts are 'utf-8' encoding including vocabulary and transcriptions.
2. Optimize error rate calculation. Mainly reduce space requirement from O(m*n) to O(min(m, n)).