伪装语音GFCC与共振峰混合特征的提取与识别

Extraction and Recognition of Mixed Features of GFCC and Formant for Camouflaged Speech

研究基于混合特征的伪装语音声纹特征提取和识别问题。分析了共振峰及GFCC这两种特征参数的提取方法,针对伪装语音对声纹识别性能的影响,提出从初始语音数据中提取基于共振峰、GFCC及其差分系数的混合特征方法,获得了鲁棒性更强、更具区分特性的说话人特征。通过和常规的基于GFCC特征的识别方法实验分析比较,所提出的基于混合特征参数的声学系统将发声特性与听觉特性结合,可以更好地体现话者的信息,具有更高的识别准确率。

The problem of voiceprint feature extraction and recognition of camouflaged speech based on mixed features is studied. The extraction methods of the formant and GFCC are analyzed. Aiming at the influence of camouflage speech on voiceprint recognition performances, a hybrid feature extraction method based on formant, GFCC and its difference coefficient is proposed from the initial speech data.A more robust and discriminative speaker feature is obtained.Through the experimental analysis and comparison with the conventional recognition method based on GFCC features, the proposed acoustic system based on mixed feature parameters combines the vocal characteristics and auditory characteristics, which can better reflect the information of the speaker and has higher recognition accuracy.