基于传统能零比和自相关函数主副峰结合的端点检测法

端点检测是语音信号处理中的一个非常重要的步骤,其准确度直接影响语音信号处理的速度和效果.传统的端点检测方法可以在高信噪比环境下准确地检测语音端点,但在低信噪比情况下,传统的端点检测特征参数不能充分描述语音信号的特征,导致端点检测效果的下降.为此,本文提出了一种对语音进行改进的多窗谱减法降噪和中值滤波减少低信噪比环境下无...

Endpoint detection method based on traditional energy zeroratio and autocorrelation function

Endpoint detection is a very important step in speech signal processing,and its accuracy has a great impact on the speed and result of speech signal processing.Traditional endpoint detection methods can accurately detect speech endpoint in high SNR environment,but in the case of low SNR,the traditional feature parameters for endpoint detection can not fully describe the characteristics of speech signal This leads to serious degradation of the effect of endpoint detection,Therefore,this paper proposes an improved multi window spectral subtraction denoising and median filtering to reduce the fluctuation of no speech segment in low SNR environment.After that,the endpoint detection method combines logarithmic energy,zero crossing rate and the ratio of main and secondary peaks of autocorrelation function.The experiment shows that the method has better accuracy and robustness than the traditional detection methods,and it has better accuracy and robustness in low SNR environment Good results of endpoint detection are obtained.