Scholarlink

This paper focuses on improving the accuracy of emotion classification models by studying the effects of augmenting the training datasets used for audio analysis, i.e., RAVDESS and TESS with additional datasets, i.e., SAVEE and CREMA-D. The study also evaluates the performances of various classifiers and techniques like SVM, Random Forest, Naive Bayes, CNN, RNN and LSTM models on the objective of emotion detection and classification. Out of all the models analyzed, CNN coupled with Bidirectional LSTM architecture was found to be most optimal due to its higher accuracy and faster training speeds as compared to other models.

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

Audio Sentiment Analysis Using Dataset Augmentation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mohammad Atif,
  • Mohd Aadil,
  • Khushi Goyal,
  • Himanshi Bharti,
  • Sonam Gupta,
  • Pradeep Gupta

摘要

This paper focuses on improving the accuracy of emotion classification models by studying the effects of augmenting the training datasets used for audio analysis, i.e., RAVDESS and TESS with additional datasets, i.e., SAVEE and CREMA-D. The study also evaluates the performances of various classifiers and techniques like SVM, Random Forest, Naive Bayes, CNN, RNN and LSTM models on the objective of emotion detection and classification. Out of all the models analyzed, CNN coupled with Bidirectional LSTM architecture was found to be most optimal due to its higher accuracy and faster training speeds as compared to other models.