根据人耳听觉特性，提出新的同步多带最大似然线性回归算法用于噪声环境下语音识别。该算法采用最大似然作为参数估计准则，利用各频带信号同步感知和噪声污染假定的方法进行语音模型补偿，有效地提高了识别系统在噪声环境下的识别性能。

Based on perception characteristic of human ear, this paper proposes synchronization multi-band maximum likelihood linear regression algorithm for robust speech recognition under noisy condition. The algorithm utilizes maximum likelihood as estimation criteria to compensate the effects of noisy condition with multi-band synchronization model and noise corruption assumption. The tests show that the proposed algorithm improves the performance of recognition system effectively.