Case Study: UT Austin achieves 96% MEG phrase classification accuracy with MathWorks MATLAB and deep learning

UT Austin Researchers Convert Brain Signals to Words and Phrases Using Wavelets and Deep Learning

UT Austin researchers partnered with **MathWorks** to tackle the challenge of building a speech-driven brain-computer interface for ALS patients, aiming to let them communicate by imagining specific phrases. Using **MATLAB** and its signal processing and machine learning tools, the team worked with MEG brain-signal data that needed heavy denoising, preprocessing, and analysis before it could be used for classification.

With **MathWorks** products including **Wavelet Toolbox, Deep Learning Toolbox, Statistics and Machine Learning Toolbox,** and **Parallel Computing Toolbox**, the researchers combined wavelet scalograms with deep neural networks to decode imagined speech. The approach achieved **96% classification accuracy**, improved on earlier SVM/ANN results of about **80%**, and cut training time by **10x** on a seven-GPU server, while enabling rapid implementation of new models and signal-processing workflows.

UT Austin

Jun Wang

UT Austin