Continuous Speech Recognition for Pocket PC based on a Small Vocabulary in Mandarin

摘要

This paper described a continuous speech recognition engine for Pocket PC (PPC) based on our previous large vocabulary continuous speech recognition technologies. Due to the limited memory resources and computing powers on PPC, we have to choose compact acoustic and language model with high share quality. Our experiments show that in small vocabulary recognition task, local computation could achieve a good performance with an applicable scheme.