摘要
描述了一个新型实时声控用户交换机语音接口的设计和实现。该系统使用数字信号处理器ＴＭＳ３２０Ｃ３１开发板在ＰＣ机ＷＩＮＤＯＷ平台下实现了与人无关连续命令语音识别，完成拨号、进行电话交换。系统使用少量控制词加上连续数字串构成了电话用户交换系统的常用命令语句，并生成相应识别文法网络（语言模型）。识别采用改进的令牌传递式Ｖｉｔｅｒｂｉ算法。研究中还开发识别系统拒识算法，在无拒识情况下命令语句中数字识别率为９８％以上，数字串（串长＜４）识别率达到９１％以上，通过加入拒识算法，字符串识别率可达９５％以上。
The design and implementation of new speech recognition system based on private automatic branch exchange (PABX) is described. Used developing board of digital signal processor (TMS320C31), speaker independent continuous speech recognition system is implemented on WINDOW platform of PC. The system uses small functional words and digits that compose different commands of telephone exchange and produce language models of recognition system. The system can make telephone exchange by speech commands over telephone channel. The recognition system uses improved token passing Viterbi algorithm, which gets good recognition accurate. Rejecting recognition algorithm is developed. In the command speech the digital recognition accuracy is over 98% and the digital string (<4 character) is over 91% without rejection and over 95% with 10% rejection of utterances.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
1998年第9期51-54,共4页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金
国家"八六三"高技术项目
