Squeeze for Sneeze: Compact Neural Networks for Cold and Flu Recognition

Abstract

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.

AI-Based human audio processing for COVID-19: A comprehensive overview

“…Albes et al. [47] addressed the memory and power consumption issues for importing a deep learning model for detecting a cold from the speech signal. They propose network pruning and quantization techniques to reduce the model size, achieving a size reduction of 95 % in Megabytes without affecting recognition performance.…”

Is Speech the New Blood? Recent Progress in AI-Based Disease Detection From Audio in a Nutshell

“…The basic feasibility of speech-based disease detection or disease/symptom severity prediction could already be demonstrated for a wide spectrum of medical conditions ranging from acute or chronic respiratory diseases, such as cold and flu ( 34 ), COVID-19 ( 24 ), or asthma ( 23 ), via psychiatric disorders, such as anxiety disorder ( 21 ), bipolar disorder ( 22 ), or depression ( 28 ), to developmental disorders, such as autism spectrum disorder ( 30 ), and neurodegenerative diseases, such as Alzheimer's disease ( 20 ) or Parkinson's disease ( 32 ). Promising results in most of the presented studies suggest that AI-based speech analysis might really have the potential to make a valuable contribution to future healthcare.…”

Audio, Speech, Language, & Signal Processing for COVID-19: A Comprehensive Overview

“…With the smartphone likely being the most convenient and available asset that every individual carries all the time, more of smartphone-based applications for detecting COVID-19 symptoms will help in controlling the virus spread. The authors of [101] have addressed the memory and power consump- The behavioural parameters such as stress, anxiety and depression needs greater attention to be paid in these days. An audio-visual emotion detection challenge organised by [105] provides a speech dataset named, "Distress Analysis Interview Corpus -Wizard of Oz (DAICWOZ)" which is a part of a larger corpus, the "Distress Analysis Interview Corpus" (DAIC) [106], A major challenge given the social distancing norms is getting the relevant and accurate speech data for developing machine learning models.…”