Comparing the Effects of Dry, Water and Gel-Based Electrodes on EEG-Based Overt Speech Classification

Recent developments in EEG based overt speech recognition have shown that speech recorded with an EEG can be classified well, however there have yet to be actual applications developed for it. This is most likely due to the EEG setup being unintuitive to the layperson. The Gel-based electrodes used in most literature are both hard and time consuming to setup. To move towards a more user friendly alternative to the current standard, this work compares Dry, Water-based and Gel-based electrodes in EEG based overt speech classification. We ran a study with 20 participants collecting EEG data of speech for five keywords. Our findings show that the Temporal muscle is most important to classification, as opposed to the Frontalis and Masseter muscle for all three electrode types. However, we were also able to show that there are no overlapping important EEG channels between the three electrode types. Finally, we found that Water-based and Dry electrodes are a suitable alternative for Gel-based electrodes.

I. Introduction

Advances in Automated Speech Recognition (ASR), have led to robust and widely used ASR systems [1] [12]. However, in the presence of background noise the audio based ASR approach is impaired [6]. EEG based overt speech recognition is an alternative ASR method, where speech is recognized using EEG measurements instead of a microphone. Using EEG has the advantage of not being as physically affected by audio background noise as an audio based paradigm, so it can still be used well in loud environments [7].