摘要
描述了单呼数字识别和连呼数字识别两种语音识别器的实现。两种识别器采用维特比算法的令牌传递版本得以实现。在有限的测试集中进行测试,单呼识别器的识别精度达100%,连呼识别器表现较差,经手工调整HMM概率后其最大识别精度为49%。
It describes the implementation of two speech recognisers, for isolated and connected digit recognition. Both recognisers are implemented using a token passing version of the Viterbi algorithm. The isolated recogniser is shown to be 100% accurate across the limited test set. The connected recogniser performs poorly with a maximum recognition accuracy of 49% after manual tweaking of the HMM probabilities.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2004年第z1期675-677,684,共4页
Chinese Journal of Scientific Instrument
关键词
维特比算法
令牌传递
隐马尔柯夫模型
概
率
Viterbi algorithm Token passing Hidden Markov models Probability