摘要
现有的基于脚步声的身份识别研究方法均采用语音声学参数作为特征,例如梅尔倒谱系数、频谱包络相似度等,这类方法的缺点是对于同一人在不同鞋类和不同地板上的脚步声识别有很大的约束性和限制性,对不同的发声机制较为敏感。提出一种新的步声特征提取方法,用步声持续时间与步声间隔时间作为识别特征,最后使用k-nn算法做识别测试。实验结果表明,该方法对于脚步声的身份识别是有效的,对于不同发声机制下的脚步声有良好的适用性。
All existing footstep-based personal identification research methods take acoustics parameters as features, for instance, Mel-Frequency Cepstral Coefficients （MFCC） and spectrum envelope similarity degree etc. Their common shortcomings are the severe constrain and limitation at recognizing the footsteps of one person with different shoes or on different floors since they are sensitive to different acoustic mechanisms. The article proposes a novel footstep feature extraction method, which takes footstep duration and interval lengths as recognizable features and finally uses k-nn algorithm for recognition test. Experiment results elaborate that the method is effective at footstep personal identification while it is well adaptive to footsteps under different acoustic mechanisms.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第1期162-164,共3页
Computer Applications and Software
关键词
脚步声身份识别
发声机制
步声间隔时间
步声持续时间k近算法
Footstep Personal identification Acoustic mechanism Footstep interval length Footstep duration Length K-nearest neighbors
