Direction of arrival estimation for multiple sound sources using\n convolutional recurrent neural network

2017/10/27 by Sharath Adavanne, Adavanne, Sharath, Archontis Politis +3 · 2 citations
Computer Science · Earth and Planetary Sciences · Engineering · #Speech and Audio Processing #Underwater Acoustics Research #Acoustic Wave Phenomena Research

paper · pdf · doi:10.48550/arxiv.1710.10059

Abstract

This paper proposes a deep neural network for estimating the directions of\narrival (DOA) of multiple sound sources. The proposed stacked convolutional and\nrecurrent neural network (DOAnet) generates a spatial pseudo-spectrum (SPS)\nalong with the DOA estimates in both azimuth and elevation. We avoid any\nexplicit feature extraction step by using the magnitudes and phases of the\nspectrograms of all the channels as input to the network. The proposed DOAnet\nis evaluated by estimating the DOAs of multiple concurrently present sources in\nanechoic, matched and unmatched reverberant conditions. The results show that\nthe proposed DOAnet is capable of estimating the number of sources and their\nrespective DOAs with good precision and generate SPS with high signal-to-noise\nratio.\n

Cited by

• Integrating Spatial and Semantic Embeddings for Stereo Sound Event Localization in Videos
• ASAudio: A Survey of Advanced Spatial Audio Research

Related

• SRP-DNN: Learning Direct-Path Phase Difference for Multiple Moving Sound Source Localization
• Direction of Arrival Estimation of Noisy Speech Using Convolutional Recurrent Neural Networks with Higher-Order Ambisonics Signals
• Localization, Detection and Tracking of Multiple Moving Sound Sources\n with a Convolutional Recurrent Neural Network
• Direct source and early reflections localization using deep deconvolution network under reverberant environment
• Improving trajectory localization accuracy via direction-of-arrival derivative estimation