Investigating the use of Deep Convolutional Neural Networks for Direction-of-Arrival Estimation on Raw Stereo Audio

This paper outlines the use of raw stereo audio as the input to a Deep Convolutional Neural Network for the purpose of Direction-of-Arrival (DOA) estimation based on source localisation over the frontal hemisphere of a 50-point Lebedev loudspeaker array using a pair of spaced stereo microphones. Results show the model is capable of a high classification accuracy with evidence of some ability to generalise to unseen data highlighting the benefits of raw audio as the input feature to models.