一种基于HMM模型的音频场景分析技术

A Method of Audio Scene Analysis Based on Hidden Markov Model

音频场景分析技术对机器自动感知环境特征具有重要的意义。考虑到Mel频率倒谱系数MFCC(Melfrequen-cycepstralcoefficient)在一定程度上可以模拟人耳的听觉感知特性,因此,提出用MFCC作为音频识别特征,通过隐马尔可夫模型进行音频场景分析的方法,该方法对7种典型场景的识别率在90%以上。

Audio scenes analysis technique plays a very important role for automatic awareness of environmental fea-tures.For the reasons of the auditory characters of human being,the MFCC(Mel frequency cepstral coefficient ),which can represent the auditory characters,is adopted as the feature of audio signal,and a method of audio scene analysis based on HMM is proposed.For typical seven audio scenes,the proposed method gets more than90%recognition rate.