Zoom out Search Issue

IEEE SIGNAL PROCESSING MAGAZINE [123] MARCH 2015
latter case). In the scenario where nonlinear speech enhancement
(noise suppression and dereverberation) was activated, three meas-
ures stood out: HASQI, PEMO-Q, and PEMO-Q-HI. Interestingly,
for the nonenhanced and enhanced cases, HASPI, a metric tailored
for intelligibility prediction, outperformed HASQI (its quality pre-
dictor counterpart) and all other metrics in terms of
.
sig
t Such
findings resonate with what was mentioned in
the section “HA: NFC Conditions” that alternate mappings of HAS-
QI’s internal parameters could be devised to reduce
f-RMSE. For
nonintrusive measures, in turn, it was found that all tested metrics
achieved insignificantly different
sig
t values in the noisy condition,
with ModA achieving the highest
sig
t and lowest f-RMSE. In the
enhanced conditions, on the other hand, only ModA achieved lev-
els above ITU-T’s “acceptability threshold.” Interestingly, in the
nonenhanced conditions (i.e., noise-alone, reverberation-alone,
and noise-plus-reverberation) ITU-T P.563 achieved reliable results
in line with those obtained with SRMR-HA and ModA. With speech
enhancement enabled, however, both P.563 and SRMR-HA perfor-
mances decreased to unacceptable levels, thus suggesting that
these two metrics are not capable of detecting and quantifying the
effects of speech enhancement artefacts on perceived quality.
These findings motivate the need for more research on the devel-
opment of innovative nonintrusive quality measures for HA
devices with nonlinear speech enhancement.
CONCLUSIONS
This article has provided a comprehensive review of 12 existing
objective quality and intelligibility prediction algorithms that have
been developed for NH and HI listeners who are users of assistive
listening devices, such as HAs and CIs. The algorithms were tested
on three common subjectively rated speech data sets: one with
subjective ratings collected from CI users in noisy and reverberant
environments, one from HA users in noisy and reverberant envi-
ronments with and without speech enhancement, and one from
HA users with NFC. The recommended metrics to be used under
each condition (nonenhanced, enhanced, NFC) were tabulated for
the two different assistive devices. In summary, for CI devices, two
measures stood out: STOI (intrusive) and SRMR-CI (nonintru-
sive). For HA with NFC, several intrusive measures attained com-
parable results, including the recently proposed PEMO-Q-HI.
None of the tested nonintrusive measures, on the other hand,
achieved acceptable results, thus leading us to explore the develop-
ment of a new metric called SRMR-HA
comp
. Finally, for HA with
speech enhancement enabled, the HASQI and PEMO-Q-HI intru-
sive measures stood out alongside ModA, a recently proposed non-
intrusive measure. It is hoped that these insights will be useful not
only for those in the assistive listening device research and devel-
opment community but also clinicians, audiologists, and patients
who wish to quickly gauge the performance of different devices
across different practical environmental conditions.
ACKNOWLEDGMENTS
Tiago H. Falk and João F. Santos acknowledge funding from the
Natural Sciences and Engineering Research Council of Canada
(NSERC) and the Fonds de Recherche du Quebec–Nature et
Technologies. Vijay Parsa and Susan Scollie acknowledge funding
from NSERC, the Oticon Foundation, and Phonak AG; James M.
Kates and Kathryn Arehart received funding from GN ReSound
and the National Institutes of Health R01 DC012289 (KA); Oldooz
Harati received funding from the National Institute of Deafness
and other Communication Disorders of the National Institutes of
Health R01 DC 007527 (PI: Philipos C. Loizou); and Rainer Huber
acknowledges funding from the German Research Foundation
(DFG) FOR-1732.
AUTHORS
Tiago H. Falk ([e-mail]) received the Ph.D. degree from
Queen’s University, Kingston, Canada, in 2009. From 2009 to
2010, he was a postdoctoral fellow at the University of Toronto. In
December 2010, he joined INRS-EMT (Montreal) as an assistant
professor. His research interests include multimedia quality meas-
urement and enhancement and human–machine interaction. He
has published over 130 papers in top-tiered journals and confer-
ences and has won four Best Paper Awards. He is a member of the
IEEE Signal Processing Society’s Speech and Language Technical
Committee, the Sigma Xi Society, and the editorial board of Jour-
nal of the Canadian Acoustical Association and Canadian Jour-
nal of Electrical and Computer Engineering. He is a Senior
Member of the IEEE.
Vijay Parsa ([e-mail]) received the Ph.D. degree in
biomedical engineering from the University of New Brunswick,
Canada, in 1996. He then joined the Hearing Health Care
Research Unit at the University of Western Ontario, where he
worked on developing speech processing algorithms for audiology
and speech language pathology applications. Between 2002 and
2007, he served as the Oticon Foundation chair in acoustic signal
processing. He is currently an associate professor jointly appointed
across the Faculties of Health Sciences and Engineering. His
research interests are in speech signal processing with applica-
tions to hearing aids, assistive listening devices, and augmentative
communication devices.
João F. Santos ([e-mail]) received his B.S.
degree in electrical engineering from the Federal University of
Santa Catarina, Brazil, in 2011 and his M.Sc. degree in tele-
communications from INRS in 2014, where he placed on the
dean’s list and was awarded the Best M.Sc Thesis Award. He is
currently pursuing his Ph.D. degree in telecommunications at
the same institute. His main research area is speech signal
processing with an emphasis in speech quality assessment and
enhancement for hearing aids and cochlear implants. He is
also interested in applications of bioinspired algorithms and
sparse representations to audio and speech processing.
Kathryn Arehart ([e-mail]) is a professor
in the Speech, Language, and Hearing Sciences Department at the
University of Colorado at Boulder. Her laboratory’s research
focuses on understanding auditory perception and the impact
hearing loss has on listening in complex auditory environments.
Current projects include the study of individual factors (cognition,
hearing loss, auditory processing) that affect the ability of older
adults to successfully use advanced hearing-aid signal processing
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
q
q
M
M
q
q
M
M
q
M
THE WORLD’S NEWSSTAND
®
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
q
q
M
M
q
q
M
M
q
M
THE WORLD’S NEWSSTAND
®
__________________
_____________
___________
__________