低信噪比下采用感知语谱结构边界参数的语音端点检测算法

发布时间：2018-02-13 11:45

本文关键词： 语音端点检测边界信息信噪比语音增强语音信号含噪语音纯净语音听觉感知参数检测结构元素 　出处：《声学学报》2014年03期 　论文类型：期刊论文

【摘要】：提出了一种采用感知语谱结构边界参数(PSSB)的语音端点检测算法,用于在低信噪比环境下的语音信号预处理。在对含噪语音进行基于听觉感知特性的语音增强之后,针对语音信号的连续分布特性与残留噪声的随机分布特性之间的不同点,对增强后语音的时-频语谱进行二维增强,从而进一步突出连续分布的纯净语音的语谱结构。通过对增强后语音语谱结构的二维边界检测,提出PSSB参数,并用于端点检测。实验结果表明,在白噪声-10 dB到10 dB的各种信噪比环境下,采用PSSB参数的端点检测算法,相对于其它端点检测算法,更有效地检测出语音的端点。在-10 dB的极低信噪比下,提出的方法仍然有75.2%的正确率。采用PSSB参数的端点检测算法,更适合于低信噪比白噪声环境下的语音端点检测。
[Abstract]:A speech endpoint detection algorithm using perceptual spectral structure boundary parameter (PSSB) is proposed, which is used to preprocess speech signal in low SNR environment. Aiming at the difference between the continuous distribution of speech signal and the random distribution of residual noise, the time-frequency spectrum of enhanced speech is enhanced in two dimensions. Through the two-dimensional boundary detection of enhanced speech spectrum structure, the PSSB parameters are proposed and used in endpoint detection. The experimental results show that, In various signal-to-noise ratio (SNR) environments with white noise of 10 dB to 10 dB, the endpoint detection algorithm with PSSB parameters is more effective than other endpoint detection algorithms. The proposed method still has an accuracy of 75.2%, and the endpoint detection algorithm based on PSSB parameters is more suitable for speech endpoint detection in low SNR white noise environment.
【作者单位】：苏州大学物理与光电·能源学部;苏州大学电子信息学院;
【基金】：国家自然科学基金(61071215,61271359,61372146) 苏州市科技发展计划(应用基础研究)(SYG201033)资助
【分类号】：TN912.3

【参考文献】