论文部分内容阅读
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 ,并基本上实时处理
The palm PC is a new, smart personal digital assistant (PDA) that currently has a soft keyboard or handwriting recognition as its primary input because it does not have a keyboard.If this platform is provided with voice-like navigation, Voice dialing and other functions, will greatly improve the human-computer interaction interface.For handheld applications such as this, combined with its slow computing speed, memory and other characteristics of the latest design of a palm core speech recognition core algorithms and implementations, including End-point detection algorithm based on time-domain energy, comprehensive judgment based on neural network multi-credibility decision-making set of extra words, feature selection and fixed-point implementation, etc.Experiments show that the selection of appropriate feature parameters, combined with fixed-point algorithm can guarantee no more than 200 A command recognition task to achieve the recognition rate of 95% or more, and basically real-time processing