Abstract:
Aiming at the problem that recognition rate is low when using Mel-Frequency Cepstral Coefficient(MFCC) to present the abnormal sounds, this paper proposes an improved feature extraction method to handle the issue, including the analysis of the characteristics of abnormal sounds and the redesign of the filter bank of MFCC. Experimental results conducted on the database of public abnormal sounds show that the method substantially outperforms MFCC feature extraction method in recognition rate and efficiency, and it can be applied in practice.
Key words:
abnormal sound,
Mel-Frequency Cepstral Coefficient(MFCC),
filter bank,
Hidden Markov Model(HMM),
feature extraction
摘要: 针对采用梅尔倒谱系数(MFCC)表征异常声音时识别率低下问题,提出获取MFCC的改进方法,包括对公共场所典型异常声音信号的特性分析和MFCC提取过程中滤波器组的重新设计。基于公共场所异常声音数据库的实验结果表明,与MFCC特征提取方法相比,该方法提高了特征参数在识别系统中的效率,具有一定的优越性和实用性。
关键词:
异常声音,
梅尔倒谱系数,
滤波器组,
隐马尔可夫模型,
特征提取
CLC Number:
LUAN Shao-wen; GONG Wei-guo. Feature Extraction of Typical Abnormal Sounds in Public Places[J]. Computer Engineering, 2010, 36(7): 208-210.
栾少文;龚卫国. 公共场所典型异常声音的特征提取[J]. 计算机工程, 2010, 36(7): 208-210.