eduzhai > Applied Sciences > Engineering >

Robust Sound Source Tracking Using SRP-PHAT and 3D Convolutional Neural Networks

  • Save

... pages left unread,continue reading

Document pages: 24 pages

Abstract: In this paper, we present a new single sound source DOA estimation andtracking system based on the well-known SRP-PHAT algorithm and athree-dimensional Convolutional Neural Network. It uses SRP-PHAT power maps asinput features of a fully convolutional causal architecture that uses 3Dconvolutional layers to accurately perform the tracking of a sound source evenin highly reverberant scenarios where most of the state of the art techniquesfail. Unlike previous methods, since we do not use bidirectional recurrentlayers and all our convolutional layers are causal in the time dimension, oursystem is feasible for real-time applications and it provides a new DOAestimation for each new SRP-PHAT map. To train the model, we introduce a newprocedure to simulate random trajectories as they are needed during thetraining, equivalent to an infinite-size dataset with high flexibility tomodify its acoustical conditions such as the reverberation time. We use bothacoustical simulations on a large range of reverberation times and the actualrecordings of the LOCATA dataset to prove the robustness of our system and itsgood performance even using low-resolution SRP-PHAT maps.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×