基于噪声分类和字典选择的语音活动检测

发布时间：2018-01-07 23:31

本文关键词：基于噪声分类和字典选择的语音活动检测　出处：《华中科技大学学报(自然科学版)》2016年12期 　论文类型：期刊论文

【摘要】：为了解决已有语音活动检测方法在噪声条件下性能急剧下降的问题,提出了一种基于稀疏编码的语音活动检测方法.该方法在训练阶段为语音和每种可能的噪声训练字典;在识别阶段首先识别环境噪声类型,然后将语音字典和对应环境噪声字典拼接形成一个大字典用于稀疏分解,最后用语音字典上的稀疏表示实现语音和非语音的判断.该方法一方面引入了噪声类型识别,可以有针对性地选择噪声字典;另一方面引入噪声开集识别机制,能识别新的噪声类型并且为之训练模型.实验结果表明所提方法比传统方法具有更强的噪声鲁棒性.
[Abstract]:In order to solve the existing voice activity detection method in noise condition performance degradation problem, proposes a voice activity detection method based on sparse encoding. In the training stage for voice and every possible noise training dictionary; first to identify environmental noise types at the recognition stage, then the speech dictionary and the corresponding environmental noise the dictionary spliced into a large dictionary for sparse decomposition, finally using sparse speech dictionary representation on Realization of speech and non speech judgment. This method introduces noise type identification, can be selected according to the noise of the dictionary; on the other hand, the noise introduced by the open set recognition mechanism, and can identify new types of noise the training model. The experimental results show that the noise robustness of the proposed method has better than the traditional method.

【作者单位】：哈尔滨理工大学计算机科学与技术学院;哈尔滨理工大学自动化学院;
【基金】：国家自然科学基金青年基金资助项目(61305001)
【分类号】：TN912.3
【正文快照】： 语音活动检测(voice activity detector)的目的是为了确定信号中语音的起始点和终止点,通常被应用于语音编码、语音增强、语音识别等语音处理系统中,具有降低语音编码速率、节省通信带宽、减少设备能耗和提高识别率等作用[1].该技术具有十分重要的理论研究和实际应用价值,一直

【相似文献】