PalmPC语音识别算法及实现

SPEECH RECOGNITION ALGORITHM DESIGN AND IMPLEMENTATION ON PALM PC

掌上型计算机 (palm PC)是一种新型、灵巧的个人数字助理 (PDA) ,由于其没有键盘 ,目前采用软键盘或手写体识别作为主要的输入手段 .如果在该平台上提供类似于语音导航、声音拨号等功能 ,将大大改善人机交互界面 .针对掌上型计算机这种应用需求 ,结合其运算速度慢、内存少等特点 ,讨论了最新设计的一个掌上型计算机语音识别核心算法及实现 ,包括基于时域能量的端点检测算法、基于神经网络的多可信度综合判决处理集外词、特征选择及定点实现等 .实验表明选择合适的特征参数 ,结合定点算法可以保证不多于 2 0 0个命令识别任务情况下达到识别率 95 %以上 。

Palm PC is a kind of novel subtle personal digital assistant (PDA). Without keyboard device,it currently adopts soft keyboard or handwriting recognition as its major input method. If functions on the platform are enhanced by providing speech navigation,voice dialing, etc, the friendly interface between human beings and machine will be ameliorated much better. In view of such kinds of application,and considering palm PC's restricted computability and limited memory, the speech recognition algorithm design and implementation on a palm PC is described in detail, including time domain energy based endpoint detection, neural network based integration of multiple confidence measures for OOV rejection, and feature selection and fix point realization. Experiments show that the recognition engine can recognize up to 200 commands in real time with less than 5% error rate.