Exploring Tradeoffs in Models for Low-latency Speech Enhancement
Sound
2018-11-20 v1 Audio and Speech Processing
Abstract
We explore a variety of neural networks configurations for one- and two-channel spectrogram-mask-based speech enhancement. Our best model improves on previous state-of-the-art performance on the CHiME2 speech enhancement task by 0.4 decibels in signal-to-distortion ratio (SDR). We examine trade-offs such as non-causal look-ahead, computation, and parameter count versus enhancement performance and find that zero-look-ahead models can achieve, on average, within 0.03 dB SDR of our best bidirectional model. Further, we find that 200 milliseconds of look-ahead is sufficient to achieve equivalent performance to our best bidirectional model.
Keywords
Cite
@article{arxiv.1811.07030,
title = {Exploring Tradeoffs in Models for Low-latency Speech Enhancement},
author = {Kevin Wilson and Michael Chinen and Jeremy Thorpe and Brian Patton and John Hershey and Rif A. Saurous and Jan Skoglund and Richard F. Lyon},
journal= {arXiv preprint arXiv:1811.07030},
year = {2018}
}