低信噪比环境下改进的倒谱距离语音端点检测算法  

Speech endpoint detection with low SNR based on improved cepstrum distance method

作　　者：张涛 章小兵 朱明星 

机构地区：[1]安徽工业大学电气与信息工程学院,安徽马鞍山243000

出　　处：《电声技术》2017年第7期108-112,125,共6页Audio Engineering

基　　金：安徽工业大学产学研基金资助重大项目(RD14206003)

摘　　要：在强背景噪声的情况下,针对传统倒谱距离法端点检测难以判断语音段起止点的问题,提出了一种基于多窗谱估计的谱减法与改进的倒谱距离语音端点检测新方法。首先对每一帧带噪信号进行多窗谱估计得到平滑功率谱,提取前导无话段平均功率谱,再利用谱减法对带噪语音信号进行减噪处理,对语音的减噪是为了更好地进行下一步的端点检测,然后对传统的倒谱距离门限阈值进行改进,得到一种改进的自适应阈值,并结合倒谱距离法进行端点检测。通过仿真实验结果表明,与传统倒谱距离端点检测算法相比,本文方法提高了低信噪比语音端点检测的精度,具有良好的鲁棒性能。Within the case of strong background noise, In order to solve the problem that the traditional cepstrum distance method is difficult to judge the start and end of the speech segment, A new method of acoustic endpoint detecting based on multi-window spectrum estimation spectral subtraction and improved cepstrum distance. The smoothed power spectrum is obtained by multi-window spectrum estimation for each frame noise signal, then, the average power spectrum of the pream- ble is extracted, besides, the noisy speech signal is denoised by spectral subtraction to better carry on the next step of the endpoint detection, Improve the traditional cepstrum distance fixed threshold, get an improved adaptive threshold, combined with cepstrum distance method for endpoint detection. The simulation results show that compared with the traditional crosstalk distance endpoint detection algorithm, this method improves the accuracy of low SNR speech endpoint detection and has good robust performance.

关 键 词：端点检测 多窗谱估计谱减法 自适应门限阈值 倒谱距离 平滑功率谱 

分 类 号：TN912.35[电子电信—通信与信息系统]