摘要
针对非平稳噪声环境下的语音活动检测问题,提出了基于在线单类SVM的自适应语音活动检测算法。该算法采用单类SVM对多种特征信息进行在线学习与综合,为非平稳背景噪声建模,并采用双层决策机制,能有效提高语音活动检测的稳健性。在语音识别系统中的实验结果表明,算法在多种噪声环境和信噪比条件下有效,并明显提高了在非平稳噪声环境下的识别率。
To improve the robustness of voice activity detection for non-stationary noises, a new algorithm based on on-line one-class SVM is proposed. This algorithm uses one-class SVM to model the noisy environment by on-line learning and integrating on multiple features. And a two-level decision mechanism which can effectively improve the robustness of speech detection is utilized in construction of the algorithm. Experimental results of speech recognition system indicate that this proposed algorithm is efficient in various noisy environments at any SNR. And the recognition performance in non-stationary noise can also be increased obviously.
出处
《深圳信息职业技术学院学报》
2008年第2期17-22,共6页
Journal of Shenzhen Institute of Information Technology
基金
深圳市科技计划项目[SZKJ0708]