Speech Emotion Recognition Based on CNN-BiGRU with a Three-Branch Parallel Fusion Architecture.

Saved in:

Bibliographic Details
Title:	Speech Emotion Recognition Based on CNN-BiGRU with a Three-Branch Parallel Fusion Architecture.
Authors:	Yu, Lie¹ lyu@wtu.edu.cn, Chen, Sijun² 3173578669@qq.com
Source:	IAENG International Journal of Computer Science. Jul2026, Vol. 53 Issue 7, p2750-2763. 14p.
Subjects:	Emotion recognition, Convolutional neural networks, Recurrent neural networks, Affective computing, Data fusion (Statistics), Human-computer interaction, Deep learning
Abstract:	Speech emotion recognition (SER) plays a pivotal role in human-computer interaction systems; however, effectively capturing both spatial and temporal features within speech signals remains a significant challenge. In this paper, a novel deep hybrid model integrating a convolutional neural network (CNN) with a bidirectional gated recurrent unit (Bi-GRU) is proposed for automatic speech emotion recognition. The CNN-BiGRU architecture employs a three-branch fusion strategy, wherein the CNN branch is responsible for spatial feature extraction, the forward GRU is utilised to model temporal dependencies, and the backward GRU is employed to capture reverse temporal information. The model was trained and evaluated on a speech emotion dataset encompassing six fundamental emotions (anger, fear, happy, neutral, sad, and surprise), with a 75:25 training-to-test split ratio being adopted. Key acoustic features were subjected to min-max normalisation prior to being input into the network. Experimental results demonstrate that a classification accuracy of 92.67% is achieved on the test set, outperforming traditional single-architecture models. Balanced performance is exhibited across all emotion categories, with high precision, recall, and F1 scores being obtained. The findings indicate that the three-branch feature fusion architecture effectively leverages complementary information from CNN and BiGRU, thereby delivering superior performance in speech emotion recognition tasks. This work provides a viable solution for practical applications in affective computing. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Computer Science is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Speech emotion recognition (SER) plays a pivotal role in human-computer interaction systems; however, effectively capturing both spatial and temporal features within speech signals remains a significant challenge. In this paper, a novel deep hybrid model integrating a convolutional neural network (CNN) with a bidirectional gated recurrent unit (Bi-GRU) is proposed for automatic speech emotion recognition. The CNN-BiGRU architecture employs a three-branch fusion strategy, wherein the CNN branch is responsible for spatial feature extraction, the forward GRU is utilised to model temporal dependencies, and the backward GRU is employed to capture reverse temporal information. The model was trained and evaluated on a speech emotion dataset encompassing six fundamental emotions (anger, fear, happy, neutral, sad, and surprise), with a 75:25 training-to-test split ratio being adopted. Key acoustic features were subjected to min-max normalisation prior to being input into the network. Experimental results demonstrate that a classification accuracy of 92.67% is achieved on the test set, outperforming traditional single-architecture models. Balanced performance is exhibited across all emotion categories, with high precision, recall, and F1 scores being obtained. The findings indicate that the three-branch feature fusion architecture effectively leverages complementary information from CNN and BiGRU, thereby delivering superior performance in speech emotion recognition tasks. This work provides a viable solution for practical applications in affective computing. [ABSTRACT FROM AUTHOR]
ISSN:	1819656X