基于DCT增强和改进谱熵的语音端点检测方法  

Speech Endpoint Detection Method Based on DCT Enhancement and Improved Spectral Entropy

作　　者：董胡[1] 钱盛友[1] 刘洋[1] 谭乔来[1] 

机构地区：[1]湖南师范大学物理与信息科学学院,湖南长沙410081

出　　处：《通信技术》2007年第11期364-365,368,共3页Communications Technology

基　　金：湖南省自然科学基金资助项目(编号:02JJY2059)

摘　　要：端点检测是语音信号处理过程中非常重要的一步,它的准确性直接影响到后面的学习及识别效果。提出了基于DCT(离散余弦变换)增强和改进谱熵的语音端点检测方法。首先,通过DCT(离散余弦变换)进行语音增强,然后再通过改进谱熵法判决去噪后语音的端点位置。仿真结果表明:此方法快速有效,具有较强的抗噪能力,特别适合低信噪比下的端点检测。The endpoint detection is an important step in speech signal processing, which directly influences the afterward speech recognition. In the paper, a speech endpoint detection method based on DCT（Discrete Cosine Transform）enhancement and improved spectral entropy is proposed. The speech is first enhanced according to DCT（Discrete Cosine Transform）,then endpoint location of denoising speech is detected by improved spectral entropy. The simulation results indicate that the proposed method candetect the endpoint exactly in the low SNR environments.

关 键 词：离散余弦变换 阈值 去噪 改进谱熵 端点检测 

分 类 号：TN912[电子电信—通信与信息系统]