%0 Journal Article
%T 基于门控复归单位(GRU)和多头注意机制的语音情感识别模型<br>A Speech Emotion Recognition Model Based on Gated Recurrent Units (GRU) and Multi-Head Attention Mechanism
%A 郭凤婵
%A 吴毅良
%A 罗序良
%A 刘翠媚
%J Artificial Intelligence and Robotics Research
%P 363-374
%@ 2326-3423
%D 2024
%I Hans Publishing
%R 10.12677/airr.2024.132038
%X 本研究提出了一种基于门控复归单位(GRU)和多头注意机制的语音情感识别模型。随着人工智能和情感计算的进步，该模型旨在分析语音信号中的情感信息，以识别说话者的情感状态，包括喜怒哀乐等各种情感表达。这一技术在情感智能、智能客服和人机交互等领域有着广阔的应用前景。本研究结合了GRU的时序信息处理能力和多头注意机制对重要特征的关注度提升，构建了一个有效且精确的语音情感识别模型。实验结果表明，此模型在IEMOCAP和Emo-DB数据集上分别实现了81.04%和94.93%的未加权准确率，相较于已有模型有显著提升。此外，该模型还展现出良好的泛化性能和可扩展性，为智能语音交互、情感计算等领域提供了可靠的技术支持。<br />
This study proposes a speech emotion recognition model based on Gated Recurrent Units (GRU) and a multi-head attention mechanism. With the advancement of artificial intelligence and affective computing, the model aims to analyze emotional information in speech signals to identify the emotional states of speakers, encompassing various expressions such as joy, anger, sadness, and others. This technology holds broad application prospects in affective intelligence, intelligent customer service, and human-computer interaction. Integrating the temporal information processing capability of GRU and the elevated attention to crucial features by the multi-head attention mechanism, an effective and precise speech emotion recognition model is developed. Experimental results demonstrate that this model achieved an unweighted accuracy of 81.04% on the IEMOCAP dataset and 94.93% on the Emo-DB dataset, showing significant improvement compared to existing models. Additionally, the model exhibits good generalization performance and scalability, providing reliable technical support for intelligent speech interaction, affective computing, and related fields.
%K 语音情感识别(SER)，门控复归单位(GRU)，多头注意机制，Bi-GRU，深度学习<br>Speech Emotion Recognition (SER)
%K Gated Recurrent Units (GRU)
%K Multi-Head Attention Mechanism
%K Bi-GRU
%K Deep Learning
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=88028