[1]周 帅,李 理,彭章君,等.基于多通道特征和混合注意力的环境声音分类[J].计算机技术与发展,2023,33(08):43-50.[doi:10. 3969 / j. issn. 1673-629X. 2023. 08. 007]
 ZHOU Shuai,LI Li,PENG Zhang-jun,et al.Environmental Sound Classification Based on Multi-channel Features and Mixed Attention[J].,2023,33(08):43-50.[doi:10. 3969 / j. issn. 1673-629X. 2023. 08. 007]
点击复制

基于多通道特征和混合注意力的环境声音分类()
分享到:

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
33
期数:
2023年08期
页码:
43-50
栏目:
媒体计算
出版日期:
2023-08-10

文章信息/Info

Title:
Environmental Sound Classification Based on Multi-channel Features and Mixed Attention
文章编号:
1673-629X(2023)08-0043-08
作者:
周 帅1 李 理12 彭章君1 黄鹏程1
1. 西南科技大学 计算机科学与技术学院,四川 绵阳 621000;
2. 四川省自主可控人工智能工程技术中心,四川 绵阳 621000
Author(s):
ZHOU Shuai1 LI Li12 PENG Zhang-jun1 HUANG Peng-cheng1
1. School of Computer Science and Technology,Southwest University of Science and Technology,Mianyang 621000,China;
2. Sichuan Autonomous Controllable Artificial Intelligence Engineering Technology Center,Mianyang 621000,China
关键词:
环境声音分类多通道特征通道注意力时频注意力混合注意力模型深度模型
Keywords:
environmental sound classification multi - channel feature channel attention time - frequency attention mixed attentionmodeldeep model
分类号:
TP391. 42
DOI:
10. 3969 / j. issn. 1673-629X. 2023. 08. 007
摘要:
环境声音分类(ESC) 已成为非常重要的研究方向,但由于环境声音种类繁多,无法进行统一表征,加之易受噪声的干扰,使得 ESC 任务变得复杂。 为了提高?
ESC 任务的识别精度,提出了基于多通道特征和混合注意力模型的分类方法。 首先,将 ESC 信号进行时频转换并使用多种滤波器提取频谱特征,将其重构
为三通道特征图。 多通道特征可以利用特征之间的互补性,弥补单一特征信息表征不足的缺点;其次,引入了一种由通道和时频注意力模块组成的混合分类
模型,通道注意力模块计算特征图并对不同通道分配权重,含有更多有效信息且对该类声音分辨较好的通道特征则会被分配更多的权重,时频注意力模块会
重点关注时域和频域中更有效的信息。 该方法可较好地抑制背景噪声,消除冗余,提高收敛速度和分类精度。 对比实验表明,在 ESC-10,ESC-50 数据集上
的识别精度分别达到了 96. 25% 和 89. 56% ,在 Ur鄄bansound8k 的数据集上达到 98. 40% 。
Abstract:
Environmental sound classification ( ESC) has become a very important research direction. However,the task of ESC becomescomplicated due to the?
variety of environmental sounds,which cannot be characterized uniformly,and the susceptibility to noise. In orderto improve the recognition accuracy of?
ESC task,a classification method based on multi-channel feature and mixed attention model isproposed. Firstly,the ESC signal is converted into time-frequency,and the spectral features are extracted by a variety of filters,which arereconstructed into a three-channel feature map. Multi-channel features?
can make use of the complementarity between features to make upfor the lack of single feature information representation. Secondly, a hybrid classification model consisting of channels and time -frequency attention modules is introduced. The channel attention module calculates the feature map and assigns weights to differentchannels. The channel features with more valid information and better resolution for this type of sound will be assigned more weights.
The time-frequency attention module will focus on more valid information in the time domain and frequency domain. The proposedmethod can suppress the background noise,eliminate the redundancy,and improve the convergence speed and classification accuracy. Thecomparison experiment shows that the recognition accuracy reaches 96. 25% and 89. 56% on ESC-10 and ESC-50 datasets respectively,and 98. 40% on Urbansound8k datasets.
更新日期/Last Update: 2023-08-10