VQ/HMM二级音节识别的研究
RESEARCH ON VQ/HMM TWO STEP CHINESE SPEECH RECOGNITION
-
摘要: HMM技术在语音识别中得到较为成功的应用,然而VQ/HMM对大词表的识别速度及识别率仍不理想.文中根据系统实现中的实时性和识别率的要求,提出了初始码本均匀法,对参加训练的各音先对其求平均,然后用各音的平均值组成初始矢量空间.并采用码本快速迭代法以及标号直方图法与HMM识别相结合的二级识别方法,提高了系统的识别率和识别速度.此方法简单,易于实时化.Abstract: Hidden Markov model(HMM) is successfully used in speech recognition, yet it is still unsatisfactory to its recognition speed and rate. An algorithm of well distributed initial code book is proposed according to the system requirement of real time and recognition rate. It improves the system recognition accuracy by using another high speed algorithm for building code book and a method of HMM two step syllables recognition. The algorithm is simple and easy to implement in real time.
下载: