Web music emotion recognition based on higher effective gene expression programming
Introduction
Music emotion recognition, as one of the main research field of music information retrieval, has been researched for many years [1], and there are many algorithms, frameworks and applications have been proposed in the literature [2], [3], recently, Yang et al. [4] proposed a harmonizing hierarchical manifolds for cross-media retrieval and presents a new framework for multimedia content analysis and retrieval based on semi-supervised ranking and relevance feedback[5]. Obtaining a good-quality model for music emotion recognition depends on many factors, like the selection of statistical methods and feature extraction, for the later, recently, Ma et al. [6] proposed a novel feature selection method and applied it to automatic image annotation. In the study, we will mainly concern about the statistical methods, although support vector machine (SVM) has been chosen as one of the best statistical methods [7], [8], it can be time-consuming. Therefore, there remains a need for an efficient method for improving the performance of music emotion recognition.
Studies show that artificial intelligence ways are efficient to find out acceptable model for music emotion recognition [9], [10], [11]. Recently, gene expression programming (GEP) [11], as the natural development of genetic algorithms [12], [13] and genetic programming [14], has been used to deal with problem of music information retrieval, especially, Yang et al. [15] used GEP to solve the problem of music emotion recognition (for MIDI music clips), and the literature suggested that GEP can improve efficiency of model for music emotion recognition.
However, over the years, many criticisms of GEP have been as follows:
- 1.
Lack of effective selection method
Roulette-wheel sampling with elitism [16] has been used in GEP for many years, to the author's knowledge, little attention has been devoted to the selection method of GEP [11], [15].
- 2.
Processing time
Until recently, there is little information available in the literature about how to decrease the processing time of GEP [11], [15].
In the study, we address all of these issues and present a higher effective algorithm, called revised gene expression programming (RGEP), to construct the model for music emotion recognition. From the experimental results, we find that the model obtained by RGEP outperforms classification accuracy of the model by GEP and takes less processing. Moreover, a fuzzy exploring method support for “emotion vector search” will be provided too.
The paper is organized as follows: In Section 2, we give a brief introduction to the methods we used in the study: GEP, RGEP, and SVM. After a detailed introduction of the experiments in Section 3, we present the experimental results and discussions in Section 4. Finally, we summarize the key conclusions of the study in Section 5.
Section snippets
Gene expression programming
GEP uses the same kind of diagram representation of genetic programming, but the entities evolved by expression tree are the expression of a genome and individuals are often copied into the next generation based on their fitness, as determined by roulette-wheel sampling with elitism, which guarantees the cloning of the best individual to the next generation [11]. Fig. 1 shows the flowchart of GEP.
As you can see from Fig. 1, GEP starts from an initial population with many genes (individuals),
Database
A total of 726 main rhythm parts (music clips with 30 s) of relevant popular songs with MP3 format was selected in the study. According to [19], we use MARSYAS [20] to retrieve 30 features from each music clip, which includes 19 of timbral texture features (means and variances of spectral centroid, roll-off, flux, zerocrossings over the texture window (8), low energy (1), and means and variances of the first five MFCC coefficients over the texture window (excluding the coefficient corresponding
Results and discussions
We selected 600 music clips as the training set and the left 86 music clips as the testing set, and we will mainly concern about the main emotion recognition. Therefore, three steps will be made in achieving the comparison results:
Step 1: Train the classifier for 100 times separately based on training set and compare the relevant processing time in GEP, RGEP and SVM;
Step 2: Conduct the experiments for many times to find the best model;
Step 3: Test the best model on testing set.
Details will be
Conclusion
In the study, we present a revised gene expression programming based on backward-chaining evolutionary algorithm to construct the model for music emotion recognition. From the experimental results, we find that the model obtained by RGEP outperforms classification accuracy of the model by GEP and takes almost 15% less processing of GEP and even half processing time of SVM, which offers a new efficient way for solving music emotion recognition problems; And because processing time is essential
Acknowledgments
This study is partly supported by the National Natural Science Foundation of China (61070075, 61004116, 61003147).
Kejun Zhang received Ph.D. in Computer Science from Zhejiang University, China. He is currently a Post-doctor in the College of Computer Science, Zhejiang University, China. His research interests lies in music information retrieval, bioinformatics, evolutionary computation and machine learning. He is a member of IEEE. He has published many research papers in various reputable journals and conference proceedings.
References (26)
An extended one-versus-rest support vector machine for multi-label classification
Neurocomputing
(2011)- et al.
Backward-chaining evolutionary algorithms
Artif. Intell.
(2006) - et al.
Are we experienced listeners? A review of the musical capacities that do not depend on formal musical training
Cognition
(2006) - X. Hu, J.S. Downie, C. Laurier, M. Bay, A. Ehmann, The 2007 MIREX audio mood classification task: lessons learned, in:...
- D. Ververidis, C. Kotropoulos, Fast sequential floating forward selection applied to emotional speech features...
- M.M. Ruxanda, B.Y. Chua, A. Nanopoulos, C.S. Jensen, Emotion-based music retrieval on a well-reduced audio feature...
- et al.
Harmonizing hierarchical manifolds for multimedia document semantics understanding and cross-media retrieval
IEEE Trans. Multimedia
(2008) - et al.
A multimedia retrieval framework based on semi-supervised ranking and relevance feedback
IEEE Trans. Pattern Anal. Mach. Intell.
(2012) - Zhigang Ma, Feiping Nie, Yi Yang, Jasper Uijlings, Nicu Sebe, Web image annotation via subspace-sparsity collaborated...
- et al.
Support vector networks
Mach. Learn.
(1995)
Emotional image and musical information retrieval with interactive genetic algorithm
Proc. IEEE
Improving speech emotion recognition using adaptive genetic algorithms
EURASIP
Gene expression programming: a new adaptive algorithm for solving problems
Complex Syst.
Cited by (21)
Musical instrument emotion recognition using deep recurrent neural network
2020, Procedia Computer ScienceSurface EMG based handgrip force predictions using gene expression programming
2015, NeurocomputingA Novel Long Short-Term Memory Network Model For Multimodal Music Emotion Analysis In Affective Computing
2023, Journal of Applied Science and Engineering (Taiwan)Study on Music Emotion Recognition Based on the Machine Learning Model Clustering Algorithm
2022, Mathematical Problems in EngineeringIntelligent Classification Model of Music Emotional Environment Using Convolutional Neural Networks
2022, Journal of Environmental and Public HealthMusic Emotion Recognition Based on Bilayer Feature Extraction
2022, Wireless Communications and Mobile Computing
Kejun Zhang received Ph.D. in Computer Science from Zhejiang University, China. He is currently a Post-doctor in the College of Computer Science, Zhejiang University, China. His research interests lies in music information retrieval, bioinformatics, evolutionary computation and machine learning. He is a member of IEEE. He has published many research papers in various reputable journals and conference proceedings.
Shouqian Sun is a professor of the College of Computer Science and Technolgy at Zhejiang University which is located in Hangzhou, Zhejiang Province of China. He is now the Director of Modern Industrial Design Institute, Zhejiang University. Since 1999 his works are concentrated in the Computer-aided Industrial Design and Conceptual Design, Applied Ergonomics and Design, Virtual Human and New Medium Design etc.