【摘要】 在连续语音识别系统中,针对强噪声环境下传统双门限语音检测方法出现的误检问题,提出了一种结合压缩感知理论和MFCC倒谱系数的端点检测算法。该算法采用Hadamard随机观测矩阵和改进的OMP重构算法对语音信号进行压缩感知与重构,利用语音信号在离散余弦基上的近似稀疏性,提取重构信号的MFCC倒谱系数来检测语音信号的端点。仿真结果表明,提出的改进算法具有较强的鲁棒性,能满足在强噪声环境下对连续语音信号进行有效端点检测的要求。更多还原

【Abstract】 In the strong noise environment, in order to solve the mistaken identified problems of continuous speech recognition in traditional double threshold speech detection methods, an endpoint detection algorithm combining compressed sensing theory and MFCC cepstrum coefficients is proposed. The method used Hadamard random observation matrix and improved OMP algorithm to compress and reconstruct the speech signal. The MFCC cepstrum coefficiens of reconstructed signals were extracted to detect the endpoints of speech signals by using the approximate sparsity of speech signals on discrete cosine basis. The simulation results show that the presented algorithm has strong robustness and can meet the requirement of effective endpoint detection for continuous speech signals in strong noise environment.更多还原