基于小波包Bark子带方差的端点检测算法

Endpoint Detection Algorithm Based on Wavelet Packet Bark Subband Variance

传统的端点检测方法大多数抗噪性不好,基于能量和短时过零率的双参数双门限端点检测在静音状态下效果较好,但是在噪声环境中性能下降.针对这个问题,使用小波包变换把信号分解成17个Bark子带,求出平均方差值,然后采用单参数双门限方法进行端点检测.实验证明,即使在-2dB的噪声环境下,该方法仍然能取得较好的端点检测效果.

The traditional endpoint detection method has poor anti-noise performance. The dual-parameter double-threshold endpoint detection based on energy and short-time zero-crossing rate works well in the mute state, but the performance is degraded in the noisy environment. To solve this problem, the wavelet packet transform is used to decompose the signal into 17 Bark subbands, and the average variance value is obtained. Then the endpoint detection is performed by the single parameter double threshold method. The experiments show that even in the noise environment of -2dB, the method can still obtain better endpoint detection effect.