Abstract: Aiming at the problem of traditional noise estimation algorithms based on minimum statistics and recursive averaging existing long time delay, this paper proposes a quick tracking noise estimation algorithm, it applies voice activity detection into subband spectral tracking to estimate the power spectrum of non-stationary noise. Simulation results show that the algorithm ensures the accuracy of noise estimation, it tracts faster than other algorithm when the noise is highly non-stationary.

Key words: speech frame, noise estimation, subband

摘要： 针对传统基于最小统计和递归平均的噪声估计算法存在较长时间延迟的问题，提出一种快速跟踪的噪声估计算法，将语音活性检测应用于子带谱跟踪来估计非稳定噪声的功率谱。仿真结果表明，该算法保证了噪声估计的精确性，当噪声为非稳定时，相比其他算法其跟踪速度更快。

关键词: 语音帧, 噪声估计, 子带

CLC Number: 

• TP301.6