Scholarlink

With the advancement of smart devices, the demand for emotion recognition in human-computer interaction is continuously increasing. As an important medium of interaction, voice carries rich emotional information, thus Speech Emotion Recognition (SER) technology holds significant importance. This paper addresses the current issue of low recognition rates and proposes a method of generating Mel-Gamma Cepstral Coefficients (MGCC) features by fusing Mel-Frequency Cepstral Coefficients (MFCC) and Gamma-Frequency Cepstral Coefficients (GFCC) and validates it with the BiLSTM-Attention model. The experimental results show that MGCC features integrate the advantages of both, enhancing recognition rates and convergence speed. Furthermore, by predicting features with the ARIMA model and replacing some zero-padding data, the feature processing is optimized, further improving recognition performance.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Speech Emotion Recognition Based on MGCC Features and ARIMA Algorithm

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hao Lou,
  • Shaoping Shen,
  • Tianle Zhang,
  • Zhibin Li

摘要

With the advancement of smart devices, the demand for emotion recognition in human-computer interaction is continuously increasing. As an important medium of interaction, voice carries rich emotional information, thus Speech Emotion Recognition (SER) technology holds significant importance. This paper addresses the current issue of low recognition rates and proposes a method of generating Mel-Gamma Cepstral Coefficients (MGCC) features by fusing Mel-Frequency Cepstral Coefficients (MFCC) and Gamma-Frequency Cepstral Coefficients (GFCC) and validates it with the BiLSTM-Attention model. The experimental results show that MGCC features integrate the advantages of both, enhancing recognition rates and convergence speed. Furthermore, by predicting features with the ARIMA model and replacing some zero-padding data, the feature processing is optimized, further improving recognition performance.