Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Toward Efficient Speech Emotion Recognition via Spectral Learning and Attention

Created by
  • Haebom

Author

HyeYoung Lee, Muhammad Nadeem

Outline

In this paper, we propose a novel 1D-CNN-based speech emotion recognition (SER) framework to capture subtle emotional changes and improve generalization performance on diverse datasets. We use Mel-Frequency Cepstral Coefficients (MFCCs) as features and leverage a 1D Convolutional Neural Network (CNN) architecture with data augmentation techniques and channel and spatial attention mechanisms to improve the performance of the model. Experimental results on various datasets (SAVEE, RAVDESS, CREMA-D, TESS, EMO-DB, EMOVO) demonstrate that the proposed method achieves high accuracy, outperforming the existing state-of-the-art performance. (SAVEE 97.49%, RAVDESS 99.23%, CREMA-D 89.31%, TESS 99.82%, EMO-DB 99.53%, EMOVO 96.39%) This suggests that the integration of advanced deep learning techniques can significantly improve the generalization performance on various datasets, and has potential for applying SER to assistive technologies and human-computer interaction in real environments.

Takeaways, Limitations

•
Takeaways:
◦
We demonstrate the effectiveness of a 1D-CNN-based SER framework utilizing data augmentation and attention mechanisms.
◦
Achieving state-of-the-art performance on diverse datasets.
◦
Presenting the applicability of SER in the field of assistive technology and human-computer interaction in real environments.
•
Limitations:
◦
Despite the high accuracy on certain datasets, the relatively low accuracy on the CREMA-D dataset suggests room for future improvement.
◦
Additional analysis and validation of the generalization performance of the method presented in the paper is needed.
◦
Further experiments considering different language and cultural backgrounds are needed.
👍
Made with Slashpage