Cohen A, Graupe D
J Biomed Eng. 1980 Apr;2(2):97-107. doi: 10.1016/0141-5425(80)90060-6.
The availability of microprocessors opens many applications for speech recognition systems where constraints on price, weight, volume and power consumption are imposed, such as in the case of a voice control system for the severely disabled. A microprocessor based speech recognition system for the voice control of wheelchair, touch-tone phone, typewriter and environmental control unit, is described. The finite word length, fixed point arithmetics and relatively slow execution time of the microprocessor introduce several difficulties in the implementation of the recognition process. Algorithms for feature extraction (mainly the autoregressive, or linear prediction coefficients), classification and training are presented. The hierarchical memory organization is discussed. Results with the ten digit set, used to activate a touch-tone phone, are given. With this set the system exhibits less than one percent substitutions and eleven percent rejections.
微处理器的出现为语音识别系统开启了许多应用场景，这些场景对价格、重量、体积和功耗都有严格限制，比如针对严重残疾人士的语音控制系统。本文描述了一种基于微处理器的语音识别系统，该系统用于控制轮椅、按键式电话、打字机和环境控制单元。微处理器有限的字长、定点运算以及相对较慢的执行时间给识别过程的实现带来了一些困难。文中介绍了特征提取（主要是自回归或线性预测系数）、分类和训练的算法。讨论了分层存储组织。给出了用于激活按键式电话的十位数集的测试结果。使用这个数集时，该系统的替换率低于1%，拒识率为11%。
