强背景噪声下语音端点检测的算法研究

摘要

多带谱熵法对语音频段进行分带处理形成新的分带谱熵函数,在低信噪比时,该方法能够更好地检测出语音,还能体现能量分布情况,应用较为广泛.多窗谱分析方法对同一数据序列用多个正交的数据窗分别求直接谱,是一种低方差、高分辨率的谱分析方法,尤其适合非线性系统中高噪声背景下弱信号、时频演变信号的分析.提出基于多窗谱及多带谱相结合的语音检测方法,仿真结果表明:改进算法较其他算法占有绝对的优势,而且性能稳定.%The multi-band entropy estimation divides the speech band into a few separate parts to form new spectral entropy functions.It is widely used as it can detect the speech signal excellently when the SNR is low,besides,it can reflect the distribution of energy.The multitaper spectrum method is a low-variance, high-resolution spectrum analysis method which weights the signal with several orthogonal data windows, and it is particularly well-suited for the diagnosis analysis of weak signals with a time-depended amplitude and frequency against a high-noise background.The speech endpoint detection is proposed based on the combination of multitaper spectrum with multi-band entropy estimation.The simulation results show that this improved algorithm has an absolute advantage over other algorithms, and its performance is always stable.