text
stringlengths
0
512
A Machine-Learning Algorithm for the Automated Perceptual
Evaluation of Dysphonia Severity
#,⁎Benjamin van der Woerd, †Zhuohao Chen, †,1Nikolaos Flemotomos, ‡Maria Oljaca, §Lauren Timmons Sund,
†,§Shrikanth Narayanan, and §Michael M
Johns, *Hamilton, Canada, and †‡§Los Angeles, California
Summary: Objectives
Auditory-perceptual assessments are the gold standard for assessing voice quality
This project aims to develop a machine-learning model for measuring perceptual dysphonia severity of audio
samples consistent with assessments by expert raters
Methods
The Perceptual Voice Qualities Database samples were used, including sustained vowel and
Consensus Auditory-Perceptual Evaluation of Voice sentences, which were previously expertly rated on a 0–100
scale
The OpenSMILE (audEERING GmbH, Gilching, Germany) toolkit was used to extract acoustic (Mel-
Frequency Cepstral Coefficient-based, n = 1428) and prosodic (n = 152) features, pitch onsets, and recording
duration
We utilized a support vector machine and these features (n = 1582) for automated assessment of
dysphonia severity
Recordings were separated into vowels (V) and sentences (S) and features were extracted
separately from each
Final voice quality predictions were made by combining the features extracted from the
individual components with the whole audio (WA) sample (three file sets: S, V, WA)
Results
This algorithm has a high correlation (r = 0.847) with estimates of expert raters
The root mean square
error was 13.36
Increasing signal complexity resulted in better estimation of dysphonia, whereby combining the
features outperformed WA, S, and V sets individually
Conclusion
A novel machine-learning algorithm was able to perform perceptual estimates of dysphonia se-
verity using standardized audio samples on a 100-point scale
This was highly correlated to expert raters
This
suggests that ML algorithms could offer an objective method for evaluating voice samples for dysphonia se-
verity
Level of Evidence
4
Key Words: Machine learning–Voice evaluation–Perceptual voice evaluation–Automation–Artificial in-
telligence
BACKGROUND
Structured voice evaluation is a critical component of as-
sessing patients with dysphonia
Comprehensive assess -
ment typically includes both perceptual and instrumental
assessments
Auditory-perceptual analysis represents the
gold standard for the assessment of dysphonia severity
It is
inexpensive and robust.1,2 This method of voice assessment
is widely accepted in clinical applications as well as research
purposes.2-4
Despite the widespread use of perceptual evaluations, it
remains a subjective assessment and raters will develop
their own internal reference standards with inherent biases,
which impact the judgment of future voice samples.5
These internal standards can vary across time and between different raters, highlighting one critique of this form of
voice assessment
, namely reliability
Through standardized
scales, such as the Consensus Auditory-Perceptual Eva-
luation of Voice (CAPE-V) tool, high levels of consistency
within and across raters can be achieved.6,7 With tools such
as this, small-scale changes from sample to sample can be
reliably detected.4,6,8 Reliability of auditory-perceptual
evaluations has been extensively researched and, when
confounding variables are controlled, they have been
proven a robust form of voice assessment.2,4,6,8,9
Expert raters are important in th
e reliability of these
assessments.2,10-12 Speech pathology assessment is a time-
limited resource and voice evaluations are limited to the
times patients can provide voice samples
Furthermore,
these assessments typically rely on in-person voice samples,
though some research suggests that remote sample collec -
tion from non-optimized settings may be adequate for
clinical assessment.13,14 These restrictions indicate a re-
source bottleneck in these evaluations
A computer-auto -
mated perceptual evaluation tool may provide an
opportunity to relieve the resource limitations and objec -
tively measure voice samples
This might allow for interval
evaluations between in-person visits, which could increase
the total number of assessments, and ultimately could
allow for within-person normative values as targets for
tracking therapeutic improvement or decline
Recent advancements in machine learning methods have
led to many medical applications, including applications Accepted for publication June 7, 2023
Journal of Voice, Vol xx, No xx, pp
xxx–xxx
0892-1997
© 2023 The Voice Foundation
Published by Elsevier Inc
All rights reserved
https ://doi.org/10.1016/j.jvoice.2023.06.006 Presented as podium presentations at the Canadian Society of Otolaryngology
Annual General Meeting 2022 in Vancouver, British Columbia, Canada and Fall
Voice Conference 2022 in San Francisco, California, USA
From the #Department of Surgery, Division of Otolaryngology—Head & Neck
Surgery, McMaster University, Hamilton, Ontario, Canada; †Department of