[4N1-GS-1] Fundamental AI, theory: algorithm:

[4N1-GS-1-02] On the linguistic representation of consonants and vowels in speech imagery EEGs

〇Tsuneo Nitta¹, Ryo Taguchi², Shuji Shinohara³, Yurie Iribe⁴, Junsei Horikawa¹, Goh Kawai (1. Toyohashi University of Technology, 2. Nagoya Institute of Technology, 3. Tokyo Denki University, 4. Aichi Prefecture University)

[[Online]]

Keywords:Brain computer interface, EEG signals, linguistic representation

In the field of neural decoding for direct communication in brain-computer interfaces (BCIs), current research has proceeded by detecting linguistic information from brain wave, or electroencephalograms (EEGs). We have proposed a model of encoding and decoding for linguistic information L(k), k = frequency. The encoding process convolves an input spectrum of random signal W(k) and L(k) and outputs an EEG spectrum X(k). The decoding process analyzes EEG spectrum X(k) using a converter of 1/L(k). Linear predictive analysis (LPA) is applied to analyze imagined speech EEGs around the Broca area. The LPA spectrum patterns are converted to line-spectra that become closer to symbolic forms. A set of vowel spectra {X(k)} is searched and reconstructed using principal component analysis (PCA) that visualizes linguistic information through eigen-vectors φ(c, m) ; c=class, m=axes number, and subspace method (SM). We trained and tested a subject-independent vowel recognizer based on a convolutional neural network (CNN). A CNN-based classifier obtained a high recognition accuracy for imagined speech vowels. In this report, consonant-specific analysis, consonantal parts labeling, and their spectra will be presented.