基于局部频谱特征与贝叶斯决策的脚步声识别  

FOOTSTEP RECOGNITION BASED ON LOCAL SPECTROGRAM CHARACTERISTICS AND BAYESIAN DECISION

作　　者：余瑶[1] 郭建敏[1] 王晅[1] 

机构地区：[1]陕西师范大学物理学与信息技术学院,陕西西安710062

出　　处：《计算机应用与软件》2015年第12期136-139,共4页Computer Applications and Software

摘　　要：针对脚步声识别系统中背景声音与噪声影响脚步声特征的提取而导致识别率明显下降的问题,根据脚步声相对背景声音在时间与频率分布上具有一定局部性的特点,提出一种声音局部频谱特征提取方法。该方法所提取的特征反映了声音主要频率成分的局部分布及其随时间的变化规律,而且对白噪声与高斯噪声的干扰有较强的鲁棒性,并且在声音采集过程中,由于与采集设备距离等因素的变化所导致的声音强度变化无关。识别过程采用贝叶斯决策理论实现步声识别。实验结果表明,该算法识别精度高于现有算法,而且对不同背景声音与环境噪声的鲁棒性明显高于现有算法。In footsteps identification system,the recognition rate is heavily degraded due to the impacts of noise and background sound on the extraction of footsteps sound. Aiming at this issue,based on the feature of footsteps sound that it is of certain locality in terms of time and frequency distributions relative to background sound, we proposed in this paper a novel method for extracting local spectrogram characteristics. The characteristics extracted by the proposed method reflect the local distributions of the principal sound frequency components and the variations rule along with the time. Moreover,they are robust to the interferences of white noise and Gaussian noise,and invariant to the variations of footsteps intensity resulted by the different distances between the examinees and the footstep recording model. The footsteps sound recognition is implemented with Bayesian decision theory in recognition process. Experimental results showed that the proposed method had higher recognition accuracy than existing algorithms,and had observably higher robustness than existing algorithms in the presence of various kinds of background sound and environmental noises.

关 键 词：脚步声 身份识别 声谱图 局部频谱特征 关键点 贝叶斯分类 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]