The authors focus on a very general low-cost acoustic/phonetic decoder,making the best use of characteristics of Chinese speech. In learning phrases, they propose a three-stage learning concept with a very simple Viterbi segmentation and counting algorithm (VSC). In recognition, fast and accurate searching is possible due to the robust partial segmentation, syllable string network (SSN) and multi-threshold pruning strategy. The paper also describes a real-time isolated word recognition system with large vocabulary based on this decoder. The system is set up on a PC-486.
展开▼