融合多分支异构特征的语音情感识别方法研究

打开文本图片集

中图分类号：TN912.34；TP183 文献标识码：A 文章编号：2096-4706（2025）21-0065-07

Research on Speech Emotion Recognition Method Based on Multi-branch HeterogeneousFeaturesFusion

RAO Wanxian （Collge ofSmartAgriculture，GuangxiScienceandTechnologyNormal University，Laibin，China）

Abstract： Speech Emotion Recognition isoneof the important directions intheresearchofspeech information processing. Basedontheexistingresearch，thispaper proposesa SpeechEmotionRecognition methodbasedonhybrid neuralnetworkand multi-branchself-atentionfusionheterogeneous features.Inthismethod，theResidualNetwork（ResNet）andtheBidirectional Long Short-Term Memory（BiLSTM） networkareusedto extract the time-frequencyand time-series featuresof spech in multiplenetworkbranches.Aftereach branch network，amulti-head Self-Atention Mechanism is introduced toenhance the contextrepresentationofeachbranchandthebranch weightsaregeneratedbythesample-levelgating mechanismtorealize theadaptiveweightedfusionofmultipleheterogeneousacousticfeatures.Finally，thefusionfeaturesareinputintothefully conectedclasificationlayertocompleteSpeechEmotionRecognition.Theexperimentalresultsshowthattherecognition accuracy of this method on ESD dataset and CASIA dataset reaches 96.57% and 83.75% respectively，which verifies the effectiveness of the proposed method.

Keywords： Speech Emotion Recognition; Attention Mechanism; heterogeneous feature fusion

0 引言

情感是一种复杂的生理和心理活动，它们是人类特有的重要性格特征，情感能力在社会推理、决策、创造等诸多活动中起着重要作用。（剩余10842字）