Návrh a vývoj zařízení pro detekci zvukových událostí pomocí strojového učení

Abstract

This bachelor thesis deals with the design and implementation of a gunshot detection system using machine learning on embedded systems. Specifically, it focuses on a convolutional neural net work trained on spectrogram data generated from audio recordings. The proposed model was implemented on the MCXN947 microcontroller, which supports hardware acceleration for neural networks. The thesis describes the process of dataset selection and preparation, model architec ture design, its optimization and quantization for the target hardware, and the subsequent final implementation.