基于特征融合的海洋哺乳动物声音识别

doi:10.16180/j.cnki.issn1007-7820.2019.05.007

摘要/Abstract

摘要：

为了提高海洋哺乳动物声音识别算法的识别率和鲁棒性,提出了一种将梅尔倒谱系数MFCC、线性倒谱系数LFCC和时域特征融合作为特征参数进行声音识别的方法。该方法通过融合不同倒谱系数以增强对不同频段的表征能力,通过融合时域特征来更全面地描述声音信息。声音样本通过基于海洋环境下的预处理、特征提取与融合后,用支持向量机进行分类识别。相对于传统算法只针对一种或几种哺乳动物进行识别,该方法在包含61种海洋哺乳动物声音的样本库中进行测试。测试结果显示该算法较传统的梅尔倒谱系数在识别率上提升了5.5%,且在海洋低信噪比环境下有更好的识别表现。

关键词: 特征融合, 海洋哺乳动物, 声音识别, 支持向量机, 倒谱系数, 时域特征

Abstract:

In order to improve the recognition rate and robustness of marine mammal sound recognition algorithm, this paper proposed a method for sound recognition by using the fusion of MFCC,LFCC and temporal features as feature parameters. This method enhanced the characterization ability of different frequency bands by fusing different cepstral coefficients and described the sound information more comprehensively by integrating the temporal features. To be specific, each continuous marine mammal recording was first preprocessed into individual syllables. Then, cepstral coefficients and temporal features were calculated from each syllable. Finally, the fused features were identified by support vector machine. Unlike traditional algorithms which only recognized few mammals, this method was tested in a sample database containing 61 marine mammal sounds. The test results showed that the proposed algorithm improved the recognition rate by 5.5% compared with the traditional MFCC, and had a better recognition performance in the low SNR environment.

Key words: feature fusion, marine mammals, sound recognition, support vector machine, cepstral coefficients, temporal features

中图分类号:

TN912.34

钟鸣拓,蔡文郁. 基于特征融合的海洋哺乳动物声音识别[J]. 电子科技, 2019, 32(5): 32-37.

ZHONG Mingtuo,CAI Wenyu. Marine Mammal Sound Recognition Based on Feature Fusion[J]. Electronic Science and Technology, 2019, 32(5): 32-37.

图/表 8

图1

图2

图4

图5

表1

图6

图7

表2

参考文献 19

[1]	David E Hanny , et al. Marine mammal acoustic detections in the northeastern Chukchi Sea, September 2007-July 2011[J]. Continental Shelf Research, 2013(67):127-146.
[2]	Nanaware S, Shastri R, Joshi Y, et al. Passive acoustic detection and classification of marine mammal vocalizations[C]. Lucknow:International Conference on Communication and Signal Processing, 2014.
[3]	André M, Van d S M, Zaugg S , et al. Listening to the Deep: live monitoring of ocean noise and cetacean acoustic signals.[J]. Marine Pollution Bulletin, 2011,63(1-4):18-26. doi: 10.1016/j.marpolbul.2011.04.038
[4]	Ibrahim A K, Zhuang H, Erdol N, et al. A New approach for north atlantic right whale upcall detection[C]. Xi’an: International Symposium on Computer,Consumer and Control, 2016.
[5]	Brown J C, Smaragdis P . Hidden Markov and Gaussian mixture models for automatic call classification[J]. Journal of the Acoustical Society of America, 2009, 125(6):EL221.
[6]	Suleman M, Tamaki Ura. Vocalization based individual classification of humpback whales using support vector machine[C]. Aberdeen:Oceans, 2007.
[7]	Dugan P J, Rice A N, Urazghildiiev I R, et al. North atlantic right whale acoustic signal processing: Part I. comparison of machine learning recognition algorithms[C]. Long Island:Applications and Technology Conference, 2010.
[8]	Zhou X, Garcia-Romero D, Duraiswami R, et al. Linear versus mel frequency cepstral coefficients for speaker recognition[C]. Hawaii:Automatic Speech Recognition and Understanding, 2011.
[9]	Harma A. Automatic identification of bird species based on sinusoidal modeling of syllables[C]. Hong Kong:IEEE International Conference on Acoustics,Speech and Signal Processing, 2003.
[10]	宋知用 . MATLAB在语音信号分析与合成中的应用[M]. 北京: 北京航空航天大学出版社, 2013.
	Song Zhiyong. Application of MATLAB in speech signal analysis and synthesis[M]. Beijing: Beihang University Press, 2013.
[11]	张帅林 . 改进的智能家居语音关键词识别算法[J]. 电子科技, 2017,30(7):5-8.
	Zhang Shuailin . Improved speech keyword spotting algorithm in smart home[J]. Electronic Science and Technology, 2017,30(7):5-8.
[12]	李艳, 成凌飞, 张培玲 . 一种基于改进谱熵的语音端点检测方法[J]. 计算机科学, 2016,43(S2):233-236.
	Li Yan , Cheng lingfei, Zhang Peiling. Speech endpoint detection based on improved spectral entropy[J]. Computer Science, 2016,43(S2):233-236.
[13]	张冬松, 马琪 . 一种基于SVM的负载识别算法[J]. 电子科技, 2017,30(8):59-62.
	Zhang Dongsong, Ma Qi . A load identification algorithm based on SVM[J]. Electronic Science and Technology, 2017,30(8):59-62.
[14]	余清清, 李应, 李勇 . 基于SVM模型的自然环境声音的分类[J]. 计算机与数字工程, 2010,38(7):1-5.
	Yu Qingqing, Li Ying, Li Yong . A SVM-based classification approach for natural sounds[J]. Computer & Digital Engineering, 2010,38(7):1-5.
[15]	SanDiego University of California. Voices in the sea[EB/OL]. ( 2015- 07- 27)[2018-04-02].
[16]	The Cornell Lab of Ornithology.Macaulay library[EB/OL]. ( 2014- 09- 28)[2018-04-02].
[17]	Gingras B, Fitch W T . A three-parameter model for classifying anurans into four genera based on advertisement calls[J]. Journal of the Acoustical Society of America, 2013,133(1):547-59. doi: 10.1121/1.4768878
[18]	Oswald M, Oswald J N, Lammers M O , et al. Integration of real time odontocete call classification algorithm into PAMGUARD signal processing software[J]. Journal of the Acoustical Society of America, 2011,129(4):2639-2639.
[19]	González-Hernández F R, Sánchez-Fernández L P, Suárez-Guerra S , et al. Marine mammal sound classification based on a parallel recognition model and octave analysis[J]. Applied Acoustics, 2017,119(3):17-28. doi: 10.1016/j.apacoust.2016.11.016

数据集	特征	分类器	训练时间/s	识别率/%±std
海洋哺乳动物音频(61种,407个样本)	MFCC	KNN	2.02	87.88%±7.89%
		RF	3.49	89.04%±4.55%
		SVM	2.57	90.07%±3.23%
	LFCC	KNN	2.10	89.74%±9.97%
		RF	3.53	91.01%±6.64%
		SVM	2.71	91.53±5.02%
	MFCC+LFCC	KNN	4.79	93.94%±5.73%
		RF	6.11	93.81%±6.76%
		SVM	5.01	94.48±4.03%
	MFCC+LFCC+ SE+SD	KNN	4.86	94.88%±7.04%
		RF	5.91	94.91%±5.95%
		SVM	5.07	95.67%±4.96%

参考文献	特征参数	原文识别种类	识别率/% ±std
[4]	MFCC+DWT	1	91.29%±5.29%
[19]	倍频程	11	89.38%±6.86%
本文算法	MFCC+LFCC +SE+SD	61	95.67%±4.96%