cnn-doa-app

Python application for real-time estimation of audio signal DOA (Direction of Arrival). This code is a part of my Master of Engineering thesis titled "Real-time Deep Neural Networks for Microphone Array Direction of Arrival Estimation".

The application runs on Raspberry Pi with a MiniDSP UMA-8 microphone array connected to it. It is capable of estimating DOA for up to 2 sources active at the same time. This estimation is done using TensorFlow Lite models trained on synthetic signals simulated using pyroomacoustics. Jupyter Notebooks used to simulate audio, preprocess it and train the models are available in my MEng-project repository.

Screenshots from the application in single-source and multi-source mode respectively are attached below: