Musi-ABC for Predicting Musical Emotions
To address the issues of insufficient accuracy and low training efficiency in general musical emotion prediction models, we propose the muSi-ABC architecture for predicting music emotions. Specifically, in the feature extraction stage of music emotions, we use a benchmark feature set to ensure that...
Main Author: | |
---|---|
Format: | Article |
Language: | English |
Published: |
IEEE
2023-01-01
|
Series: | IEEE Access |
Subjects: | |
Online Access: | https://ieeexplore.ieee.org/document/10197391/ |
_version_ | 1797746107194802176 |
---|---|
author | Jing Yang |
author_facet | Jing Yang |
author_sort | Jing Yang |
collection | DOAJ |
description | To address the issues of insufficient accuracy and low training efficiency in general musical emotion prediction models, we propose the muSi-ABC architecture for predicting music emotions. Specifically, in the feature extraction stage of music emotions, we use a benchmark feature set to ensure that the extracted music emotion features adhere to standardization. In the prediction stage, we introduce the muSi-ABC architecture which first utilizes a 2D-ConvNet (two dimensional-Convolutional Neural network) to extract partial critical features in music emotions. Then, the BiLSTM (Bi-directional Long Short Term Memory) neural network is employed to learn contextual sequential information of past and future music emotions from the obtained partial critical features. Furthermore, the SA (Self-Attention) module is applied to obtain the complete critical features highly relevant to music emotions, thereby improving prediction accuracy and training efficiency. Through ablation experiments conducted at different time term lengths, the roles of ConvNet model and SA module, as well as the advantages of the proposed muSi-ABC architecture over other ablated models in terms of training efficiency and prediction accuracy, are verified. Additionally, it is observed that representing music emotions using long term feature information for the same song can enhance prediction accuracy. Finally, contrast experimental results demonstrate that the proposed architecture outperforms other benchmark methods in terms of prediction accuracy. Moreover, it is validated that the outlier points contained in the music emotions features extracted based on the benchmark feature set help discover the variations trends of music emotions. |
first_indexed | 2024-03-12T15:32:21Z |
format | Article |
id | doaj.art-4cfafea1ea0c4a45822b95c562dab129 |
institution | Directory Open Access Journal |
issn | 2169-3536 |
language | English |
last_indexed | 2024-03-12T15:32:21Z |
publishDate | 2023-01-01 |
publisher | IEEE |
record_format | Article |
series | IEEE Access |
spelling | doaj.art-4cfafea1ea0c4a45822b95c562dab1292023-08-09T23:01:42ZengIEEEIEEE Access2169-35362023-01-0111794557946510.1109/ACCESS.2023.330004210197391Musi-ABC for Predicting Musical EmotionsJing Yang0https://orcid.org/0009-0009-1855-4654Conservatory of Music, Qilu Normal University, Jinan, ChinaTo address the issues of insufficient accuracy and low training efficiency in general musical emotion prediction models, we propose the muSi-ABC architecture for predicting music emotions. Specifically, in the feature extraction stage of music emotions, we use a benchmark feature set to ensure that the extracted music emotion features adhere to standardization. In the prediction stage, we introduce the muSi-ABC architecture which first utilizes a 2D-ConvNet (two dimensional-Convolutional Neural network) to extract partial critical features in music emotions. Then, the BiLSTM (Bi-directional Long Short Term Memory) neural network is employed to learn contextual sequential information of past and future music emotions from the obtained partial critical features. Furthermore, the SA (Self-Attention) module is applied to obtain the complete critical features highly relevant to music emotions, thereby improving prediction accuracy and training efficiency. Through ablation experiments conducted at different time term lengths, the roles of ConvNet model and SA module, as well as the advantages of the proposed muSi-ABC architecture over other ablated models in terms of training efficiency and prediction accuracy, are verified. Additionally, it is observed that representing music emotions using long term feature information for the same song can enhance prediction accuracy. Finally, contrast experimental results demonstrate that the proposed architecture outperforms other benchmark methods in terms of prediction accuracy. Moreover, it is validated that the outlier points contained in the music emotions features extracted based on the benchmark feature set help discover the variations trends of music emotions.https://ieeexplore.ieee.org/document/10197391/Predicting musical emotionslong term dependencypartial critical featurescomplete critical points |
spellingShingle | Jing Yang Musi-ABC for Predicting Musical Emotions IEEE Access Predicting musical emotions long term dependency partial critical features complete critical points |
title | Musi-ABC for Predicting Musical Emotions |
title_full | Musi-ABC for Predicting Musical Emotions |
title_fullStr | Musi-ABC for Predicting Musical Emotions |
title_full_unstemmed | Musi-ABC for Predicting Musical Emotions |
title_short | Musi-ABC for Predicting Musical Emotions |
title_sort | musi abc for predicting musical emotions |
topic | Predicting musical emotions long term dependency partial critical features complete critical points |
url | https://ieeexplore.ieee.org/document/10197391/ |
work_keys_str_mv | AT jingyang musiabcforpredictingmusicalemotions |