装甲车内背景下的语音端点检测的改进

An Improved Method of Voice Activity Detection in Armored Vehicles

以模拟装甲车内部环境噪声对带噪语音进行端点检测,旨在通过设计良好的端点检测算法来提高后续语音增强的效果。从实战角度出发,语音端点检测的需求为较高的准确性和实时性以及鲁棒性。结合语音信号特性,采取以谱减法作为预处理的基础上通过能熵比对带噪语音进行端点检测,通过实验仿真的方式证明其可行性。

The VAD is made to the noise speech through simulating the inside environment of armored vehicle, in order to improve the effect of subsequent speech enhancement. From the practical point of view, the VAD demands the higher accuracy and timeliness and robustness. Hence, the spectral subtraction as the pretreatment is used to achieve the VAD with the energy to entropy ratio and proves its feasibility through the way of simulation experiment.