摘要
为了提高现有谱相减方法处理低信噪比语音信号的效果, 利用时间序列信号的短时分形维数对低信噪比语音信号的端点检测方法进行了研究, 提出了一种基于端点检测的谱相减语音增强方法, 给出了其原理及具体算法。仿真实验结果表明, 该方法简单可靠, 有效可行, 当语音信号信噪比降到 -5dB时仍然有效; 与已有方法相比, 具有更广泛的实用性。
To improve existing spectral subtraction methods on low SNR(Signal-to-Noise Ratio) speech signal processing, a kind of detection of noise-speech endpoint on low SNR speech signal through application of time sequence short time fraction dimension is proposed.A spectral subtraction speech enhancement method based on endpoint detection is put forward.Principle and detail method are given.By simulation to self-recorded speech signal,it turns out that the method is simple and workable.Compared with other methods,when speech signal SNR is reduced to -5 dB, the method is still workable.So the method has wider applicability.
出处
《吉林大学学报（信息科学版）》
CAS
2005年第2期139-142,共4页
Journal of Jilin University（Information Science Edition）
关键词
分形维
语音端点检测
语音增强
谱相减
fraction dimension
speech endpoints detect
speech enhancement
spectral subtraction
