基于子带能量线性映射的噪声中端点检测算法

研究噪声环境下的语音端点检测问题。在低信噪比下 ,虽然噪声和语音的频谱分布不同 ,但是传统语音检测算法使用的时域能量没有描述能量在各频域子带的分布 ,对于语音和噪声没有很好的区分性。以前提出的基于时间 -频率的能量参数利用频域的限带能量加上时域能量来进行噪声中的语音检测。但是它们选择频带的依据是语音信号的高能量子带 ,而没有考虑噪声的子带能量分布。该文提出的语音检测方法同时考虑语音和噪声的频域能量分布 ,采用线性映射的方法将 Mel滤波器组的子带能量特征空间映射到噪声和语音最有区分性的一维子空间 ,得到新的特征参数 EL MBE进行语音检测。实验结果表明 ,在噪声环境下基于线性映射的能量参数比时域能量 ,基于时间 -频率的能量有更好语音检测性能。

Robust word boundary detection through linear mapping of the sub-band energy in noisy environments

This paper analyzes word boundary detection in the presence of noise. At low signal to noise ratios, speech and noise have different energy distributions in the frequency domain, even though the energy distribution in the time domain is similar for speech and noise. Time frequency energy have been used for speech detection in noisy environments. But the frequency bands used by previous methods are high energy speech bands which do not consider the noise energy distribution. In this paper, the Mel scale band energy is linearly mapped into a one dimensional sub space, which preserves the speech and noise separability as much as possible. The energy in the mapped sub space is called the Linear Mapping Sub Band Energy ( E LMBE ). Experiments show that E LMBE outperforms the time domain energy and the time frequency energy when used for word boundary detection in noisy environments.