On Loss Functions for Deep-Learning Based T60 Estimation
Yuying Li, Yuchen Liu, Donald S. Williamson
Abstract
Reverberation time, T <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">60</inf> , directly influences the amount of reverberation in a signal, and its direct estimation may help with dereverberation. Traditionally, T <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">60</inf> estimation has been done using signal processing or probabilistic approaches, until recently where deep-learning approaches have been developed. Unfortunately, the appropriate loss function for training the network has not been adequately determined. In this paper, we propose a composite classification- and regression-based cost function for training a deep neural network that predicts T <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">60</inf> for a variety of reverberant signals. We investigate pure-classification, pure-regression, and combined classification-regression based loss functions, where we additionally incorporate computational measures of success. Our results reveal that our composite loss function leads to the best performance as compared to other loss functions and comparison approaches. We also show that this combined loss function helps with generalization.
BibTeX
@inproceedings{icassp2021_onlossfunctionsf,
title = {On Loss Functions for Deep-Learning Based T60 Estimation},
author = {Yuying Li and Yuchen Liu and Donald S. Williamson},
booktitle = {ICASSP 2021},
year = {2021}
}