Direction of Arrival Estimation of Moving Sound Sources using Deep Learning

Sound source localization is an important task for several applications and the use of deep learning for this task has recently become a popular research topic. While nearly all previous work has focused on static sound sources, in this paper we evaluate the performance of a deep learning classification system for localization of moving sound sources and we evaluate the effect of different hyperparameters and acoustic conditions. A feedforward neural network is used to estimate the direction of arrival of moving sound sources, with Short Time Fourier Transform input features. Diverse synthetic datasets are generated to represent different acoustic conditions, and hyperparameters are tested to determine which combination results in better direction-ofarrival detection for moving sources. We evaluate the performance of the different combinations in terms of precision and recall, in a multi-class multi-label classification framework, and we find that (1) the number of frequency bins and the reverberation time have a significant effect for localizing high-speed sources, and (2) precision and recall decay slowly at low speeds while dropping sharply at high speeds.

Automated summary

In this paper, we evaluate the performance of a deep learning classification system for localization of moving sound sources, and investigate the effect of different hyperparameters and acoustic conditions. The study demonstrates the significance of hyperparameters and acoustic conditions in localizing high-speed sources.