基于分形理论的语音端点检测

Study of voice activity detection based on the fractal theory

为提高语音端点检测（VAD）在较低信噪比（〈10dB）下的准确率，提出一种基于短时分形雏数的改进算法。结合语音信号的特点，对2种常用的语音信号分形雏数计算方法进行了比较和选择，同时采用动态跟随门限值实现语音端点的自适应检测。试验结果表明：对于信噪比6～10dB的带噪语音，此方法可以实现整段语音的检测，而且具有一定的噪声鲁棒性，系统运行期间能够自适应调整门限值以适应环境噪声的变化，提高了VAD算法的准确率。

A reformative algorithm, based on the short-time fractal dimension, is given to improve VAD （voice activity detection） in low signal-to-noise ratio （SNR） environments. Two ways to calculate the fractal dimensions of speech signals in common were compared, and one was selected to fit the speech signals. At the same time, a dynamically updated threshold was applied to adaptively detect the speech segments in noisy speech. The experimental results showed that the method could realize the detection of speech segment in low signal-to-noise ratio （SNR at 6-10dB） and had a good noise robustness.