Research Article
Research on speech emotion recognition based on multi-feature fusion
@INPROCEEDINGS{10.4108/eai.24-2-2023.2330662, author={Zhiqiang Huang and Mingchao Liao}, title={Research on speech emotion recognition based on multi-feature fusion}, proceedings={Proceedings of the 2nd International Conference on Engineering Management and Information Science, EMIS 2023, February 24-26, 2023, Chengdu, China}, publisher={EAI}, proceedings_a={EMIS}, year={2023}, month={6}, keywords={emotion recognition; multi-feature; cbam}, doi={10.4108/eai.24-2-2023.2330662} }- Zhiqiang Huang
Mingchao Liao
Year: 2023
Research on speech emotion recognition based on multi-feature fusion
EMIS
EAI
DOI: 10.4108/eai.24-2-2023.2330662
Abstract
In the field of emotion recognition, speech emotion recognition research is very prevalent now, and the information contained in a single feature has a limit. To address the issue of low identification accuracy caused by a single feature, we provide an audio multi-feature fusion approach that fully fuses feature information and extracts information for recognition in several dimensions in this study. Initially, pre-processing and feature extraction are conducted on audio files, and the Mel-spectrogram and multi-F feature sets are extracted. The Mel-spectrogram feature map is then fed into the Convolutional Block Attention Module (CBAM) for higher dimensional feature mapping, and the output results are cascaded with multi-F before being fed into the fully connected layer and SoftMax layer to complete the classification with an accuracy of 81.5% on IEMOCAP datasets.