3G系统中背景噪声消除算法的研究

对适用于3G终端系统单麦克风（single microphone）情况的最小均方误差估计（minimum mearesquare error estimation．简称MMSE）噪声消除算法进行了研究，比较了传统MMSE噪声消除算法与经过改进的MMSELSA（Log-Spectral Amplitude MMSE-STSA）噪声消除算法的性能和算法复杂度，其目的是从算法消除噪声的效果和最后硬件实现的难易度两方面来验证哪种算法更适合实际3G系统的应用。论文所讨论的MMSE-LSA噪声消除算法对传统MMSE算法的改进之处主要是在谱增益的计算和语音激活检测（VAD）技术两方面，其中对谱增益计算的改进改变了传统贝赛尔函数实现的算法，使实际的应用更加容易。而对VAD检测算法的改进，使含噪语音的语音停顿周期检测更加准确，从而提高了噪声消除的性能。最后通过主观和客观的检测方法，证明了后一种算法的优越性。

Research on background noise cancellation algorithm in 3G system

The research on MMSE （minimum mean-square error estimation） NC （noise cancellation） algorithm, which has been fitted for single microphone NC in 3G terminal system, and the comparison between MMSE-LSA （Log-Spectra MMSE-STSA）NC algorithm and the original MMSE NC algorithm in performance and complexity of the algorithm have been made. And the aim is to fmd the better NC algorithm for 3G system. In the article the original MMSE algorithm is improved by MMSE-LSA NC algorithm in the two aspects of spectral gain computation and VAD algorithm： the improvement on spectral gain computation changes the original Bessel function algorithm to make the application easier in practice; the improvement on VAD algorithm makes the detection of the period of audio stop more precise to enhance the performance of NC. The subjective and objective evaluations prove that the MMSE-LSA （Log-Spectral Amplitude MMSE-STSA）NC algorithm is superior to the original MMSE NC algorithm.