仿生听觉技术在声目标特征提取中的应用

Application of Biomimetic Technology to Feature Extraction from Acoustic Objects

为提高战场声目标识别技术,从仿生听觉技术原理出发,研究声目标的发声机理和人耳的听觉特性,通过与语音声学特征比较,分别研究了时域、频域、小波域和倒谱域具有代表性的声特征参数。通过矢量量化技术对不同声特征、特征矢量不同分量和不同特征组合的识别效果做了比较,实验表明:多种声目标条件下MFCC识别率为96.8%,远高于其他特征,表明其在差异性较大的目标特征中更具有区分度。

In view of principle of biomimetic technology, feature extraction from acoustic targets can influence the performance of whole sys- tem to enhance recognition performance. Several kinds of time domain, frequency domain and cepstrum domain were researched, including linear prediction cepstrum coefficient, mel-frequency cepstrum coefficient（MFCC） , dynamic acoustic feature and wavelet feature and etc. The recognition results of components of different feature for acoustic target system were given. The results show that recognition rate based on MFCC is higher than others reaching 96.8% under different object.