Audio Window Classification
Edge-based audio event classification using VGG on Raspberry Pi.
Overview
This was an early Technical University of Munich (TUM) course project aimed at performing live event classification on audio streams.
We accomplished this by fine-tuning the VGG variant (released on AudioSet in 2017) on the ESC-70 dataset. To achieve real-time inference on edge devices, the model was compressed via TensorFlow Lite and demonstrated running live on a Raspberry Pi equipped with a microphone.