Abstract: In order to monitor, diagnose and identify the machinery or equipment, the audio signal is used as the monitoring means, and the Vector Quantization（VQ） algorithm is introduced, also mechanical equipment’s Discrete Hidden Markov Model（DHMM） is established. The mechanical equipment audio parameter extracted is MFCC, and the code book is produced using Linde-Buzo-Gray（LBG） algorithm. The Baum-Welch algorithm’s simplest scaling factor form based on multiple sequences is deduced. Meanwhile logarithmic form of Viterbi algorithm is used. Various forms of HMM model are compared through the experiments, and the suitable audio HMM model form for mechanical equipment is chosen. The experiments on 22 kinds of audio signals, with the recognition accuracy rate of more than 97%, prove the validity of the method.

Key words: Mel Frequency Cepstrum Coefficient（MFCC）, Vector Quantization（VQ）, Linde-Buzo-Gray（LBG） algorithm, Discrete Hidden Markov Model（DHMM）, audio recognition, equipment monitoring

摘要： 为了对现场机械或设备进行监控、诊断和识别，以音频为监控手段，引入矢量量化（VQ）算法并建立机械设备音频的离散隐Markov模型（DHMM）。特征参数采用MFCC，码书设计采用Linde-Buzo-Gray（LBG）算法；推导出Baum-Welch算法参数重估的多观察序列的最简标定形式；分析了多种HMM类型，提出了适合机械设备音频的HMM。实验在22种音频中进行，识别准确率在97%以上，证明了方法的有效性。

关键词: Mel倒谱系数（MFCC）, 矢量量化, LBG算法, 隐马尔科夫模型, 音频识别, 设备监控