Characteristic Sequence Analysis of Giant Panda Voiceprint

Frontiers in Physics

Peng Chen¹  Rong Hou¹  Shaoxiang Hu²  Zhiwu Liao³ 

[1] Chengdu Research Base of Giant Panda Breeding, Sichuan Key Laboratory of Conservation Biology for Endangered Wildlife, Chengdu, China;School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu, China;School of Computer Science, Sichuan Normal University, Chengdu, China;

关键词: MFCC;    long-range dependent;    individual recognition;    voiceprint;    Gaussian mixture model;   

DOI  :  10.3389/fphy.2022.839699

来源: DOAJ

【 摘 要 】

By analyzing the voiceprint characteristics of giant panda’s voice, this study proposes a giant panda individual recognition method based on the characteristics of the composite Mel composite frequency cepstral coefficient (CMFCC) and proves that the characteristic sequence of the CMFCC has long-range dependent characteristics. First, the MFCC (Mel composite frequency cepstral coefficient) with a low frequency resolution is obtained by the Mel filter bank; then, the inverse Mel frequency cepstral coefficient (IMFCC) features of giant panda calls are extracted. The CMFCC characteristic sequence of giant panda voice composed of the MFCC and IMFCC improves the resolution of high- and low-frequency resolution characteristics of giant panda voice. Finally, the first-order difference characteristic parameters of the MFCC are integrated to obtain the difference characteristics between frames. Through experiments, the improvement of the system recognition effect is verified, and the recognition accuracy meets the theoretical expectation.

【 授权许可】

Unknown