基于双门限算法的端点检测改进研究

Research and Improvement on Endpoint Detection Based on Dual-threshold Algorithm

端点检测技术是语音识别中的一个重要步骤,其准确性直接影响语音信号处理的效率,近年来对语音信号端点检测技术的研究一直是语音信号处理的热点。传统的双门限端点算法在实验室环境下检测结果较为准确,但在复杂信道环境下,正确率大大下降。针对传统双门限算法的局限性,论文提出了一种基于二级判决的改进方法。该算法采用短时振幅来构造过零"带",增强复杂环境下过零率的抗干扰性。同时增设辅音能量阈值和疑似辅音阈值,解决清辅音丢失以及无声段混入问题。实验结果表明,该方法在复杂信道条件下端点检测正确率有较大提高。

The technology of endpoint detection which has been the focus of speech processing in recent years is an importantstep of speech recognition,and it has a great influence on speech processing,speech recognition and speaker identification etc. Forclear speech,The traditional dual-threshold algorithm is accurate,however,it？？s greatly reduced while in the complex channel envi？ronment. Aiming at the limitations of the traditional dual-threshold algorithm,an improved method based on two sentences is pro？posed in this paper. In order to enhance the anti interference of zero rate in the complex environment,the short term amplitude isused to construct the zero band. At the same time,the energy thresholds of the consonants and suspected consonants are used tosolve the problem of the loss of the voiceless consonants and the mixing of the silent voice segment. Experimental results prove thatthe proposed method can improve the endpoint accuracy obviously,although in the background of complex channel.