IRO Journals
IRO Journal on Sustainable Wireless Systems

EEG based Imagined Speech Classification using CWT and ResNet-50-D Architecture

Open Access
Published: 27 July, 2026
Pages: 221-235
  • Immanuel G. , Immanuel G.
    Division of Computer Science and Engineering, Karunya Institute of Technology and Sciences, Coimbatore, India
    Division of Computer Science and Engineering, Karunya Institute of Technology and Sciences, Coimbatore, India
  • Shirly S. Shirly S.
    Division of Computer Science and Engineering, Karunya Institute of Technology and Sciences, Coimbatore, India
    Division of Computer Science and Engineering, Karunya Institute of Technology and Sciences, Coimbatore, India
view PDF
view PDF

How to Cite

G., Immanuel, and Shirly S. 2026. “EEG Based Imagined Speech Classification Using CWT and ResNet-50-D Architecture”. IRO Journal on Sustainable Wireless Systems 8 (3): 221-35. https://doi.org/10.36548/jsws.2026.3.007.
ACM ACS APA ABNT Chicago Harvard IEEE MLA Turabian Vancouver AMA

Download Citation

Endnote/Zotero/Mendeley (RIS) BibTeX

Keywords

Brain–Computer Interface
Electroencephalography
Imagined Speech Classification
Continuous Wavelet Transform
ResNet-50-D
Deep Learning
Time–Frequency Analysis

Abstract

Imagined speech recognition is considered a novel BCI technique where the human brain can directly communicate with the external world without the need for speech production. Nevertheless, decoding the imagined speech using EEG data is problematic due to the non-stationary properties, signal-to-noise ratio issues, and subject-dependent nature of EEG. The current paper introduces a classification method for imagined speech using EEG signals based on the combination of Continuous Wavelet Transform and the ResNet-50-D deep learning architecture. Firstly, EEG data from the ASU Imagined Speech EEG Dataset are preprocessed to eliminate noise and artifacts. Then, the preprocessed EEG data are converted into the time–frequency representation using CWT with Morlet wavelet function, thus, allowing effective representation of time and frequency components of neural activity. The resulting representations are used as input data for ResNet-50-D. The designed scheme is tested on four different hypothetical speech types, namely vowels, short words, long words, and short-long word combinations. From the experiments, accuracies of classification achieved are 72.4%, 74.0%, 77.5%, and 76.4% correspondingly. The comparative study of the proposed approach with previous works shows the high efficiency of the combination of time-frequency features representation and residual deep learning. The results show that the proposed CWT-ResNet-50-D approach is an efficient and reliable method for decoding imagined speech from EEG and can be further used in BCI systems.

References

  1. Nguyen, Chuong H., George K. Karavas, and Panagiotis Artemiadis. "Inferring Imagined Speech Using EEG Signals: A New Approach Using Riemannian Manifold Features." Journal of neural engineering 2018, vol. 15, no. 1: 016002.
  2. Kamble, Ashwin, Pradnya Ghare, and Vinay Kumar. "Machine-Learning-Enabled Adaptive Signal Decomposition for a Brain-Computer Interface Using EEG." Biomedical Signal Processing and Control 2022, vol 74: 103526.
  3. Mohan, Anand, and R. S. Anand. "Classification of Imagined Speech Signals Using Functional Connectivity Graphs and Machine Learning Models." Brain Topography 2025, vol 38, no. 2: 25.
  4. Mohan, Anand, and R. S. Anand. "Wavelet Augmented Phase Coherence Features for EEG-Based Imagined Speech Classification." IEEE Sensors Letters 2025, vol. 9, no. 8: 1-4
  5. Saha, Pramit, and Sidney Fels. "Hierarchical Deep Feature Learning for Decoding Imagined Speech from EEG." In Proceedings of the AAAI Conference on Artificial Intelligence 2019, vol. 33, no. 01, 10019-10020.
  6. Panachakel, Jerrin Thomas, A. G. Ramakrishnan, and T. V. Ananthapadmanabha. "A Novel Deep Learning Architecture for Decoding Imagined Speech from EEG." arXiv preprint 2020, arXiv:2003.09374.
  7. Panachakel, Jerrin Thomas, Ramakrishnan Angarai Ganesan, and T. V. Ananthapadmanabha. "Common Spatial Pattern-Based Data Augmentation Technique for Decoding Imagined Speech." In 2021 IEEE International Conference on Electronics, Computing and Communication Technologies (CONECCT) IEEE 2021, 1-5.
  8. Cohen, Michael X. "A Better Way to Define and Describe Morlet Wavelets for Time-Frequency Analysis." NeuroImage 2019, vol 199: 81-86.
  9. Phoommanee, Nonpawith, Peter J. Andrews, and Terence S. Leung. "Deep Learning-Based Phase Recognition in Anterior Nasal Endoscopy for Handheld Devices." In 2025 Research, Invention, and Innovation Congress: Innovative Electricals and Electronics (RI2C), IEEE, 2025, 50-53.
  10. He, Tong, Zhi Zhang, Hang Zhang, Zhongyue Zhang, Junyuan Xie, and Mu Li. "Bag of Tricks for Image Classification with Convolutional Neural Networks." In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 2019, 558-567.
  11. Singh, Abhiram, and Ashwin Gumaste. "Interpreting Imagined Speech Waves with Machine Learning Techniques." arXiv preprint 2020, arXiv:2010.03360.
  12. Arizona State University (ASU). "Imagined Speech EEG Dataset." Arizona State University, Tempe, AZ, USA.
  13. Available: IEEE DataPort: https://ieee-dataport.org/open-access/imagined-speech-dataset