摘要
为了改进噪声环境下的语音增强效果,充分抑制背景噪声,有效消除残留"音乐噪声",本文通过MATLAB仿真测试,对目前广泛使用的功率谱减法、维纳滤波器、最大似然短时谱幅度(STSA)估计以及最小均方误差STSA估计等语音增强算法进行了性能比较,在深入理解最大似然STSA估计算法的基础上,提出了一种针对实时应用的改进型短时谱幅度(M-STSA)估计器,实验结果显示本文提出的语音增强算法物理意义明确,复杂度较低,在低信噪比情况下能有效抑制背景噪声,在高信噪比时又能减小语音的畸变.同时,本文还提出了一种新的先验信噪比估计算法,较好地消除了残留"音乐噪声".
To improve the effect of speech enhancement in noisy background, a new short-time spectral amplitude(STSA) estimator based on maximum likelihood STSA (ML-STSA) estimation is proposed and then compared with other widely used estimators which are based on Wiener filtering, "power spec- tral subtraction" (PSS) algorithm, ML-STSA estimation and minimum mean-square error STSA (MMSE-STSA) estimation. In this paper, the modified maximum likelihood STSA (MML-STSA) esti- mator is also derived under uncertainty of speech presence in the noisy observations based on modeling speech and noise spectral components as statistically independent Gaussian random variables. Evaluation results show that the proposed MML-STSA estimator not only can suppress the residual noise effectively and reduce "musical noise" phenomena in low SNR as MMSE-STSA estimator,but also can reduce the distortion of the voice in high SNR.
出处
《四川大学学报(自然科学版)》
CAS
CSCD
北大核心
2013年第6期1261-1269,共9页
Journal of Sichuan University(Natural Science Edition)
基金
国家自然科学基金(51275323)