基于短时对数谱估计MMSE的语音增强算法研究  

Study of Speech Enhancement Based on MMSE Short Time Log-spectral Amplitude Estimator

作　　者：袁伟军[1] 刘珩[1] 

机构地区：[1]中国农业大学信息与电气工程学院,北京100083

出　　处：《电声技术》2007年第10期59-62,65,共5页Audio Engineering

摘　　要：结合语音激活检测(VAD)技术对短时对数谱估计最小均方误差(LSA-MMSE)语音增强算法进行了改进。通过实验表明,LSA-MMSE增强算法在消除背景噪声、增加语音清晰度和提高语音自然度等方面比谱减法更加有效。To improve the performance of communication system and clear the noise especially the background noise in real environment in noisy speech, a method of LSA-MMSE（Log-Spectral Amplitude-Minimum Mean Square Error） combined with advanced VAD speech enhancement is proposed. Compared with the spectral subtraction, results show that the method of LSA-MMSE combined with VAD speech enhancement is more effective than the spectral subtraction in terms of noise suppression and speech signal distortion.

关 键 词：语音增强 短时谱估计 最小均方误差 

分 类 号：TN912.35[电子电信—通信与信息系统]