Implementation of Taiwanese Large Vocabulary Continuous Speech Recognition System Based on Kaldi Asr Toolkit

In this project of speech recognition, we tried to use the Kaldi speech recognition engine, targeting at Taiwanese speech (the Taiwanese Minnan language), based on a well transcribed read-aloud speech database to do the task of large-vocabulary continuous speech recognition. Another terminology for this task is so called speech-to-text for the Taiwanese language. The data base used here was recorded sponsored by Taiwan government (the Education Department of the Execution Yuan). It was downloadable from the Internet. It has been processed in Academia Sinica and Change Gung University and thus was named as “TwESC” Speech recognition database. This project will aim at achieving the syllable error rate as low as 10% and the characher error rate as low as 10 % under reasonable constraint of the language model.

Project ID:PB10703-1483
External Project ID:MOST106-2221-E182-077