复杂环境中多音节语音自适应端点检测的实现  

Realization of Adaptive Endpoint Detection for Multisyllable Speech Sound in Complex Environment

作　　者：郑丹丹 秦会斌[1] ZHENG Dan-dan;QIN Hui-bin(Institute of Electron Devices ＆ Application,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)

机构地区：[1]杭州电子科技大学新型电子器件与应用研究所,浙江杭州310018

出　　处：《计算机仿真》2018年第7期154-157,236,共5页Computer Simulation

摘　　要：对嘈杂环境下的语音信号传输真实性的检测,能够降低语音的失真度,有效保证语音的高质量传输。对语音信号传输真实性的检测,需要利用短时能量更新语音端点的判决阈值,进行自适应校验。传统方法通过语音信号的特征参数及其演变形式,提取语音阈值,但忽略了对其进行校验。提出基于短时信噪比的自适应阈值和自适应判决的语音端点检测算法,结合双门限判决法,根据自适应化短时能量动态更新端点判决阐值,辅以短时过零率和自适应判决校验,最终利用F-measure评价语音信号端点检测结果。实验结果表明,提出的自适应端点检测算法在不同环境中的平稳噪声和非平稳噪声中均能有效检测出语音中有话段和无话段之间的端点,且算法准确性和鲁棒性明显优于其它传统算法。To detect the authenticity of speech signal transmission in noisy environment can reduce the distortion of speech sound and effectively guarantee the high quality transmission of speech sound. The detection of authenticity of speech signal transmission needs to use short-time enetgy to update the decision threshold of speech endpoint and perform the adaptive check. Traditionally, the method extracts the threshold value of speech through feature parameter of speech signal and its evolution form, but the verification is ignored. Therefore, we put forward an algorithm to de- tect the speech endpoint of adaptive threshold and adaptive decision based on short time signal-to-noise ratio. Com- bined with the method of double threshold decision, the endpoint decision threshold was dynamically updated based on adaptive short-time energy. With short-time zero-crossing rate and adaptive decision check, F-measure was used to evaluate the endpoint detection resuh of speech signal. Following conclusion can be drawn from experimental re- suits. The proposed adaptive endpoint detection algorithm can effectively detect the endpoints between speech segment and non-speech segment from stationary noise and non-stationary noise in different environments. Meanwhile, the accuracy and robustness of algorithm is obviously better than that of other traditional algorithms.

关 键 词：语音信号 信号传输 真实性检测 

分 类 号：TN912.3[电子电信—通信与信息系统]