摘要
通过对噪音和语音频谱的分析,针对航空背景噪声的特性,提出一种用于语音增强的新的噪声估计算法。通常的噪声估计一般利用语音端点检测方法,取噪声段的谱平均值作为待估计的噪声谱,但该方法在信噪比较低时性能下降严重。笔者提出的基于频率段能量比的噪音谱估计方法,不依赖于语音端点检测而直接由语音帧来估计噪音谱,通过计算一帧语音中各频率段中能量比,以判断该帧是否含有语音来修正噪声谱估计的计算因子。算法提高了谱减法的适用范围,还在一般谱相减方法的基础上提出了改进的谱相减算法。
A new noise estimation algorithm under aerospace noise feature for speech enhancement is proposed. The noise estimation are usually based on voice activity detection, but this kind of methods deteriorates quickly when the SNR is low. The new noise estimation which is based on energy proportion lies in different frequency fragment and is independent of the voice activity detection. An improved non-linear spectral subtraction algorithm is also put forward. The application range of the spectral subtraction is expanded by the new noise estimation method.
出处
《电声技术》
2006年第11期59-62,共4页
Audio Engineering
关键词
谱减法
噪音估计
语音增强
非平稳噪声
spectral subtraction
noise estimation
speech enhancement
non-stationary noise
