Sensor Signal Processing: Raw Audio to Machine Learning for command classification

An end-to-end audio signal processing pipeline for voice-controlled gaming using Python. This project captures spoken commands ('up', 'down', 'left', etc.) and processes them through a custom feature engineering workflow based on the Fast Fourier Transform (FFT). Linear Discriminant Analysis (LDA) is used to create a 2D feature space for classification. The performance of k-NN, SVM, and Decision Tree models were compared, with the best model being integrated into a final live demo that controls a Pac-Man game in real-time.

This research provides an experimental analysis of delamination, a critical failure mode in fiber-reinforced plastic composites. To investigate its impact on structural integrity,
E-glass/epoxy laminates were manufactured with embedded artificial defects of varying shapes, sizes, and interlayer locations.

The mechanical properties of these samples were then evaluated using a three-point bending test. The results demonstrate a clear and quantifiable reduction in flexural stiffness, showing that larger defects and those located closer to the composite's center cause the most significant degradation in performance.