Automatic Soundscape Affect Recognition Using A Dimensional Approach

被引:21
作者
Fan, Jianyu [1 ]
Thorogood, Miles [1 ]
Pasquier, Philippe [1 ]
机构
[1] Simon Fraser Univ, SIAT, Burnaby, BC V5A 1S6, Canada
来源
JOURNAL OF THE AUDIO ENGINEERING SOCIETY | 2016年 / 64卷 / 09期
关键词
PLEASURE;
D O I
10.17743/jaes.2016.0044
中图分类号
O42 [声学];
学科分类号
070206 ; 082403 ;
摘要
Soundscape affect recognition is essential for sound designers and soundscape composers. Previous work demonstrated the effectiveness of predicting valence and arousal of soundscapes in responses from one expert user. Based on this, we present a method for the automatic soundscape affect recognition using ground truth data collected from an online survey. An analysis of the corpus shows that participants have a high level of agreement on the valence and arousal of soundscapes. We generate a gold standard by averaging users' responses, and we verify the corpus by training stepwise linear regression models and support vector regression models. An analysis of the models shows our system obtains better results than the previous study. Further, we test the correlation between valence and arousal based on the gold standard. Last, we report an experiment of using arousal as a feature for predicting valence and vice versa.
引用
收藏
页码:646 / 653
页数:8
相关论文
共 28 条
[1]  
[Anonymous], 2010, ISMIR
[2]  
[Anonymous], 2007, P INT C EXH NOIS CON
[3]  
[Anonymous], P 4 INT C COMP CREAT
[4]  
[Anonymous], 2010, Proc. ismir
[5]  
[Anonymous], FRONTIERS PSYCHOL
[6]  
[Anonymous], 16 EUR SIGN PROC C L
[7]  
[Anonymous], INT C AC
[8]  
[Anonymous], 2013, NIME
[9]  
[Anonymous], 1998, TUTORIAL SUPPORT VEC
[10]  
Aucouturier J. J., 2007, P 10 INT C AC MADR S