摘要
提出一种基于快速噪声估计的MMSE语音增强算法,实验表明这种算法比起谱相减法和基于语音短时对数谱的最小均方误差(MMSE-LSA)算法能更显著地提高算法的客观性能,在非平稳噪声环境中能快速估计出变化的噪声功率谱。
A fast noise-estimation algorithm used in the speech enhancement algorithm of MMSE is studied.Experiments results prove that this algorithm can dramatically improve objective performance compared with spectrum subtraction algorithm and MMSE-LSA algorithm,and it can estimate the noise spectrum quickly in the non-stationary noise environment.
出处
《计算机工程与应用》
CSCD
北大核心
2007年第22期113-114,119,共3页
Computer Engineering and Applications
基金
国家自然科学基金(the National Natural Science Foundation of China under Grant No.60472094)
山西省高校科技开发项目(No.197-12909083)。
关键词
噪声估计
最小均方误差
语音增强
noise estimation
Minimum-Mean-Square-Error (MMSE)
speech enhancement