摘要
可靠的语音端点检测算法是稳健语音识别系统所必须的。针对现有算法在噪声环境下的稳健性问题,提出了基于单类SVM(Support Vecfor Machine)的端点检测算法。通过对多特征信息进行在线学习与综合,以及采用双层决策机制,有效提高了语音检测的稳健性。实验表明,算法在多种噪声环境和信噪比条件下有效,明显提高了语音识别系统在噪声环境下的识别率。
Reliable endpoint detection is crucial to the performance of the robust speech recognition system. To improve the robustness of this kind of algorithm in noisy environment, a new algorithm based on one-class SVM (support vector machine) is proposed. This algorithm not only uses an online learning and integrating strategy on multiple features but also introduces a two-level decision mechanism, which helps effectively improve the robustness of speech detection. Experimental results show that this proposed algorithm is efficient in various noisy environments at any SNR and the performance of speech recognition system is also obviously increased.
出处
《深圳信息职业技术学院学报》
2006年第4期19-24,共6页
Journal of Shenzhen Institute of Information Technology
关键词
语音识别
端点检测
语音活动检测
单类支持向量机
speech recognition
endpoint detection
voice activity detection
one-class SVM (support vector machine)