一种改进的基于子带谱熵的语音激活检测方法

An improved voice activity detection algorithm based on band-partitioning spectral entropy

语音信号的激活检测(voice activity detection,VAD)是变速率语音编码的关键技术,用来检测通信时是否有语音片存在。在有噪环境下对语音信号的激活检测是非常重要而困难的。对传统子带谱熵算法进行了改进,提出了一种能够用于语音激活检测的新算法。该算法利用语音谱熵和噪声谱熵分布的不同,将信号的数字特征(方差、均值等)与传统子带谱熵相结合,用于区分语音段和非语音段。计算机仿真结果表明,在高斯白噪声环境下,改进后的子带谱熵算法能很好地区分说话人的语音段和非语音段,在某种程度上解决了传统语音激活检测算法结构复杂、参数难调、易受噪声影响等问题。

Voice activity detection （VAD） is the key technology of variable rate speech coding to distinguish the speech sentences from non-speeches. Voice activity detection is important but difficult in noisy environments. Hence, an improved spectral entropy algorithm was introduced into the voice activity detection. By using the different distribution feature between the speech entropy and non-speeches entropy, the spectral entropy was combined with the statistics feature to distinguish the speech sentences from non-speeches. Simulation results show that this algorithm successfully distinguishes speech sentences from non-speeches under the white noise. To some extent, the improved algorithm overcomes the traditional algo-rithms＇shortcomings, such as structural complexity, not easy to adjust the parameters, and sensitive to noise.