TY - GEN
T1 - Squeeze for sneeze
T2 - 21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020
AU - Albes, Merlin
AU - Ren, Zhao
AU - Schuller, Björn W.
AU - Cummins, Nicholas
N1 - Publisher Copyright:
Copyright © 2020 ISCA
PY - 2020
Y1 - 2020
N2 - In digital health applications, speech offers advantages over other physiological signals, in that it can be easily collected, transmitted, and stored using mobile and Internet of Things (IoT) technologies. However, to take full advantage of this positioning, speech-based machine learning models need to be deployed on devices that can have considerable memory and power constraints. These constraints are particularly apparent when attempting to deploy deep learning models, as they require substantial amounts of memory and data movement operations. Herein, we test the suitability of pruning and quantisation as two methods to compress the overall size of neural networks trained for a health-driven speech classification task. Key results presented on the Upper Respiratory Tract Infection Corpus indicate that pruning, then quantising a network can reduce the number of operational weights by almost 90 %. They also demonstrate the overall size of the network can be reduced by almost 95 %, as measured in MB, without affecting overall recognition performance.
AB - In digital health applications, speech offers advantages over other physiological signals, in that it can be easily collected, transmitted, and stored using mobile and Internet of Things (IoT) technologies. However, to take full advantage of this positioning, speech-based machine learning models need to be deployed on devices that can have considerable memory and power constraints. These constraints are particularly apparent when attempting to deploy deep learning models, as they require substantial amounts of memory and data movement operations. Herein, we test the suitability of pruning and quantisation as two methods to compress the overall size of neural networks trained for a health-driven speech classification task. Key results presented on the Upper Respiratory Tract Infection Corpus indicate that pruning, then quantising a network can reduce the number of operational weights by almost 90 %. They also demonstrate the overall size of the network can be reduced by almost 95 %, as measured in MB, without affecting overall recognition performance.
KW - Cold
KW - Compact Neural Networks
KW - Computational Paralinguistics
KW - Flu Recognition
KW - Pruning
KW - Quantisation
UR - http://www.scopus.com/inward/record.url?scp=85098156244&partnerID=8YFLogxK
U2 - 10.21437/Interspeech.2020-2531
DO - 10.21437/Interspeech.2020-2531
M3 - Conference contribution
AN - SCOPUS:85098156244
SN - 9781713820697
T3 - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
SP - 4546
EP - 4550
BT - Interspeech 2020
PB - International Speech Communication Association
Y2 - 25 October 2020 through 29 October 2020
ER -