【摘要】 本文介绍了一个基于定点数字信号处理器ADSP2181的语音控制器,着重讨论了其语音特征参数计算方法。该控制器采用Mel尺度倒谱系数（MFCC）作为基本识别参数,文中首先讨论了其标准浮点实现方法,然后结合其算法特点和定点DSP的结构,提出了一种优化的快速定点实现算法。测试结果表明,定点算法计算时间比浮点算法大大降低,计算256点（32ms）语音数据12阶MFCC参数只需1.35ms。极大地降低了DSP的负荷。更多还原

【Abstract】 Introduced in this paper is a speech controller based on fixed-point DSP ADSP2181,which adopts mel-scale frequency cepstrum coefficient as principal speech recogniti0n parameter.We discuss its standard floating-point scheme and then, taking into account the features of its algorithm and the structure of the fixed-point DSP, put forward an optimized, fast fixed-point scheme. The testing shows that the fixed-point scheme consumes much less time than the floatingpoint scheme. Twelve MFCC coefficients of a frame of 256 point samples (32ms) cost onIy 1.35ms.Compared with 12.20ms for the floating-point scheme, the load of the DSP is considerably reduced.更多还原