Sorry, you need to enable JavaScript to visit this website.

facebooktwittermailshare

Multi-scale Octave Convolutions for Robust Speech Recognition

Abstract: 

We propose a multi-scale octave convolution layer to learn robust speech representations efficiently. Octave convolutions were introduced by Chen et al [1] in the computer vision field to reduce the spatial redundancy of the feature maps by decomposing the output of a convolutional layer into feature maps at two different spatial resolutions, one octave apart. This approach improved the efficiency as well as the accuracy of the CNN models. The accuracy gain was attributed to the enlargement of the receptive field in the original input space. We argue that octave convolutions likewise improve the robustness of learned representations due to the use of average pooling in the lower resolution group, acting as a low-pass filter. We test this hypothesis by evaluating on two noisy speech corpora - Aurora-4 and AMI. We extend the octave convolution concept to multiple resolution groups and multiple octaves. To evaluate the robustness of the inferred representations, we report the similarity between clean and noisy encodings using an affine projection loss as a proxy robustness measure. The results show that proposed method reduces the WER by up to 6.6% relative for Aurora-4 and 3.6% for AMI, while improving the computational efficiency of the CNN acoustic models.

up
0 users have voted:

Paper Details

Authors:
Joanna Rownicka, Peter Bell, Steve Renals
Submitted On:
16 May 2020 - 8:58am
Short Link:
Type:
Presentation Slides
Event:
Presenter's Name:
Joanna Rownicka
Paper Code:
SPE-P8.11
Document Year:
2020
Cite

Document Files

ICASSP2020_JRownicka_slides.pdf

(14)

Subscribe

[1] Joanna Rownicka, Peter Bell, Steve Renals, "Multi-scale Octave Convolutions for Robust Speech Recognition", IEEE SigPort, 2020. [Online]. Available: http://sigport.org/5374. Accessed: Jul. 15, 2020.
@article{5374-20,
url = {http://sigport.org/5374},
author = {Joanna Rownicka; Peter Bell; Steve Renals },
publisher = {IEEE SigPort},
title = {Multi-scale Octave Convolutions for Robust Speech Recognition},
year = {2020} }
TY - EJOUR
T1 - Multi-scale Octave Convolutions for Robust Speech Recognition
AU - Joanna Rownicka; Peter Bell; Steve Renals
PY - 2020
PB - IEEE SigPort
UR - http://sigport.org/5374
ER -
Joanna Rownicka, Peter Bell, Steve Renals. (2020). Multi-scale Octave Convolutions for Robust Speech Recognition. IEEE SigPort. http://sigport.org/5374
Joanna Rownicka, Peter Bell, Steve Renals, 2020. Multi-scale Octave Convolutions for Robust Speech Recognition. Available at: http://sigport.org/5374.
Joanna Rownicka, Peter Bell, Steve Renals. (2020). "Multi-scale Octave Convolutions for Robust Speech Recognition." Web.
1. Joanna Rownicka, Peter Bell, Steve Renals. Multi-scale Octave Convolutions for Robust Speech Recognition [Internet]. IEEE SigPort; 2020. Available from : http://sigport.org/5374