Latent Gaussian Activity Propagation: Using Smoothness And Structure To Separate And Localize Sounds In Large Noisy Environments

Authors:
Daniel Johnson Harvey Mudd College
Daniel Gorelik Harvey Mudd College
Ross E Mawhorter Harvey Mudd College
Kyle Suver Harvey Mudd College
Weiqing Gu Harvey Mudd College
Steven Xing Intel Corporation
Cody Gabriel Intel Corporation
Peter Sankhagowit Intel Corporation

Introduction:

The authors present an approach for simultaneously separating and localizingmultiple sound sources using recorded microphone data.

Abstract:

We present an approach for simultaneously separating and localizingmultiple sound sources using recorded microphone data. Inspired by topicmodels, our approach is based on a probabilistic model of inter-microphonephase differences, and poses separation and localization as a Bayesianinference problem. We assume sound activity is locally smooth across time,frequency, and location, and use the known position of the microphones toobtain a consistent separation. We compare the performance of our methodagainst existing algorithms on simulated anechoic voice data and find that itobtains high performance across a variety of input conditions.

You may want to know: