一种基于巴克域噪声估计的语音增强算法

为了在保证语音增强算法性能的同时,降低算法复杂度,提出了一种巴克域最小统计量控制递归平均噪声估计算法.将带噪信号在巴克域进行分解并进行最小统计量分析,基于此最小统计量控制噪声的递归平均估计.算法基于听觉模型,充分利用巴克带内频带间的相关性,具有较好的噪声跟踪估计性能.该算法复杂度低,适用于常见语音增强方法.仿真结果表明,基于该噪声估计的语音增强可以有效地抑制噪声,增强后语音失真较小,在低信噪比条件下能够有效改善语音编码合成后的语音质量.

Speech enhancement with noise estimation in bark domain

An improved minimum controlled recursive averaging noise estimation approach in bark domain was proposed to keep the enhancement performance with lower complexity.First,the noisy speech was transformed into bark domain and the minimum was tracked.Then,the minimum was applied to controling the smoothing factor of the recursive averaging in noise estimator.The proposed method,based on perceptual model,makes full use of the correlation among frequency bins in the same bark band to track noise better.Furthermore,the approach has low complexity.The simulation results show that the proposed method can suppress noise effectively with low speech distortion,and the synthesis speech of 2400bps MELP(mixedexcitation linear predictive) speech coder with this speech enhancement has better speech quality in adverse environment.