摘要: |
语音检测是语音信号处理的前端,利用长时谱能量差异特征的语音检测无法区分突
发噪声和语音,掺杂着突发噪声的语音信号会对语音处理系统带来不良影响。提出了一
种基于长时谱能量差异特征和基音比例特征相结合的语音检测方法,该方法的优点是,在利
用长时谱能量差异特征基础上引入基音比例特征,从而有效减少了将信号中突发噪声误判为
语音的错误。实验显示,该算法能够在多种信噪比环境下取得很好的检测结果。 |
关键词: 语音信号处理 语音检测 长时谱能量差异 基音比例 突发噪声 |
DOI: |
|
基金项目: |
|
Improved voice activity detection based on long-term spectral divergence and pitch ratio features |
MENG Yi-ming,OU Zhi-jian |
() |
Abstract: |
Voice Activity Detection (VAD) is the front-end of speech processing and
the VAD algorithm which uses long-term spectral divergence (LTSD) feature can′
t discriminate abrupt noise from speech. The speech signal with abrupt noise will
adversely affect the speech processing system. This paper proposes a VAD algori
thm which combines LTSD feature and pitch ratio feature. The advantage of the al
gorithm is that by introducing pitch ratio feature, it can effectively reduce th
e false alarms of taking abrupt noise as speech. Experimental results show that
the algorithm achieves good performance for VAD under various signal-to-noise ratios. |
Key words: speech processing voice activity detection long-term spectral divergence pitch
ratio burst noise |