solderneer / nexus-silent-speech

A novel approach to enhancing silent speech recognition systems with a multiplexed star electrode array, developed for my final year dissertation at University College London (UCL). Includes datasets, code, and electronics CAD files.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Nexus, a session-independent silent speech interface

A novel approach to session-independent silent speech recognition systems with a multiplexed star electrode array, developed for my final year dissertation at University College London (UCL). Includes datasets, code, and electronics CAD files.

Overview

This repository houses my final year dissertation project at University College London (UCL), focused on developing a novel multiplexed star electrode array to address the challenges of cross-session variability in silent speech recognition systems. This innovative approach leverages configurable electrode orientations to significantly enhance the signal-to-noise ratio (SNR) of surface electromyography (sEMG) signals, leading to improved recognition accuracy across sessions.

Project Objectives

  • To investigate the impact of cross-session variability on the accuracy of silent speech recognition systems.
  • To design, develop, and evaluate a multiplexed star electrode array capable of adjusting electrode orientations for optimal muscle-electrode alignment.
  • To compare the performance of the novel star electrode array against traditional electrode configurations in terms of SNR and silent speech recognition accuracy.

Results Summary

The project successfully demonstrated the potential of the star electrode array to achieve superior muscle-electrode alignment, resulting in a significant increase in SNR compared to control configurations. Most notably, the array displayed a remarkable 41% relative improvement in cross-session classification accuracy, underscoring its effectiveness in mitigating the challenges posed by session-dependent variability.

Contents

  • datasets/: Contains the sEMG datasets collected during the project, across multiple sessions and with different electrode configurations.
  • code/: Includes all code developed for signal acquisition, signal processing, feature extraction, and classification tasks.
  • cad/: Houses the electronics CAD files for the star electrode array and control circuitry, enabling replication and further research.
  • docs/: Additional project documentation, including project report and presentation slides.

Contact

For any inquiries or collaboration proposals, please contact Shan at sudhar393@gmail.com.

About

A novel approach to enhancing silent speech recognition systems with a multiplexed star electrode array, developed for my final year dissertation at University College London (UCL). Includes datasets, code, and electronics CAD files.


Languages

Language:Jupyter Notebook 84.1%Language:C 12.9%Language:Python 2.3%Language:Dockerfile 0.5%Language:CMake 0.3%