一种基于定点DSP的语音识别算法实现

介绍了一种基于定点DSP的特定人的孤立词语音识别系统，着重讨论了算法中参数快速提取的实现．该算法采用基于动态规划（DP）技术的动态时间弯折算法（DTW），采用VAD两极端点检测方案以提高端点检测的精度，以Mel尺度倒谱系数为语音识别参数．首先讨论了标准浮点实现方法，然后结合算法特点和定点DSP的结构，采用了一种优化的快速定点实现方法．测试结果表明，定点算法时间比浮点算法大大降低，使系统的性能得到了提高．

The Realization of A Fixed-point Speech Recognition Algorithm Based on DSP

The paper introduces a specific isolated word speech recognition systems based on fixed-point DSP,focusing on achieving the rapid extraction algorithm parameters.The algorithm is based on dynamic programming(DP) technology Dynamic Time Warping algorithm(DTW),by using VAD bipolar endpoint detection programmers to improve the accuracy of endpoint detection,and MFCC for speech recognition parameters.The paper firstly discusses realizing methods of standard floating-point.Then combining with its fixed-point DSP...