Abstract: Music disturbance reduces recognition rate of the word command in the car word command recognition system.Because of its the larger eigenvalue spread and the smaller spectrum evenness，the algorithm convergence speed of the music signal is slower than that of speech signal so that traditional adaptive cancelling algorithm cannot complete cancel the music disturbance and recognition rate of the word command cannot be improved largely.To overcome this problem，a pre-whitening adaptive filter is used to reduce the eigenvalue spread and increase the spectrum evenness.In addition，in order to improve the convergence rate of the adaptive cancelling algorithm double adaptive filtering algorithm is adopted to cancel the music disturbance together with the pre-whitening adaptive filtering.The experimental results show notably improved correct recognition rate of the word command after music disturbance cancellation and pre-whitening improves recognition rate.

摘要： 在车载命令词识别系统中，背景音乐的播放降低了命令词识别率。而音乐信号因自相关矩阵特征值扩散度较大和谱平坦度较小在算法收敛速度上比语音信号慢，以至于传统的自适应抵消算法很难将音乐干扰消除干净，保证不了命令词识别率。为了解决这一问题，文中引入了预白化自适应滤波器来减小其自相关矩阵特征值扩散度和增大谱平坦度，并将此方法结合双滤波自适应算法，用来消除车内的背景音乐，以提高车载命令词识别系统的识别率。实验结果表明，经过背景音乐消除，命令词识别率有明显的提高，并且预白化也提高了识别率。