Zoom out Search Issue
IEEE SIGNAL PROCESSING MAGAZINE [122] MARCH 2015
insignificantly different
sig
t values in the nonenhanced case,
resulted in significantly higher
f -RMSE values. Such finding
shows the importance of short-time processing for CI users.
Interestingly, while PESQ and PEMO-Q have been shown to be
highly correlated with subjective quality ratings of NH listeners in
a number of telephony applications, poor performance was
obtained for CI users, particularly under speech enhancement.
For the nonintrusive measures, the SRMR-CI measure achieved
the best results with performance levels in-line with those
obtained with STOI, but with the advantage of not requiring a ref-
erence signal. In fact, when both noisy and enhanced conditions
were considered overall, the SRMR-CI metric outperformed STOI
across all four performance criteria. By incorporating CI process-
ing percepts into the original SRMR measure, significant gains
could be observed. Generally, the findings observed here resonate
with those reported in the literature showing the importance of
spectral envelopes for CI intelligibility.
HA: NFC CONDITIONS
For users of HAs with frequency lowering strategies, PEMO-Q-HI
and STOI attained insignificantly different
sig
t and f-RMSE
results. The HASQI measure, in turn, resulted in the highest
sig
t ,
but achieved a significantly higher
f-RMSE than the two afore-
mentioned metrics. This higher error may be a result of the range
of conditions used during training of the internal parameter (i.e.,
noise, linear, and nonlinear terms) mapping available in the
HASQI. Notwithstanding, given the burgeoning popularity of such
nonlinear frequency compression schemes for HI listeners with
severe to profound high frequency sensorineural hearing loss, our
results suggest that users have a few reliable intrusive metrics to
choose from. On the other hand, the tested nonintrusive measures
were not capable of correctly characterizing the perceptual arte-
facts caused by NFC in HA users. For example, none of the metrics
surpassed the correlation threshold of 0.8 established by ITU-T
during the competition that resulted in the P.563 Recommenda-
tion [20]. These findings motivate the need for more research on
the development of innovative nonintrusive quality measures for
frequency-lowering strategies. As an exploratory test, the modula-
tion energy thresholding and modulation filter bank compression
strategies implemented in the SRMR-CI metric (see the section
“SRMR-CI and SRMR-HA”) were tested on the original SRMR and
SRMR-HA metrics and significant improvements
.p 0051
^h
could be observed (e.g., .080
sig
t = and f-RMSE .468= with the
so-called SRMR-HA
comp
). In fact, these newly obtained results were
in-line with some of the intrusive metrics, such as PEMO-Q, and
suggest that further improvements may be obtained with nonin-
trusive measures.
HA: NOISY AND ENHANCED CONDITIONS
Finally, for HA users in complex listening environments compris-
ing noise, reverberation, and noise-plus-reverberation, it was
observed that all intrusive measures achieved insignificantly differ-
ent
sig
t
and
f
-RMSE values (with the exception of HASPI, in the
0 0.2 0.4 0.6 0.8 1
0
20
40
60
80
100
PESQ
Quality
0 0.2 0.4 0.6 0.8 1
0
10
20
30
40
50
60
70
80
90
ModA
(a) (b)
Quality
Not Enhanced
Enhanced
Not Enhanced
Enhanced
[FIG3] Scatterplots of subjective quality versus objective scores for condition-averaged data points obtained from the (a) PESQ and
(b) ModA metrics for the HA reverberation/enhancement quality database.
[TABLE 4] A SUMMARY OF RECOMMENDED OBJECTIVE
METRICS FOR DIFFERENT CONDITIONS. THE METRICS IN
BOLD REPRESENT THOSE THAT ACHIEVED HIGHEST
sig
t
AND LOWEST
f
-RMSE. THE METRIC SRMR-HA
comp
CORRE-
SPONDS TO AN EXPLORATORY MEASURE DESCRIBED IN
THE SECTION “HA: NFC CONDITIONS.”
CI HA
CONDITION INTRUSIVE NON-
INTRUSIVE
INTRUSIVE NONINTRUSIVE
COMBINED STOI, NCM SRMR-CI PESQ, STOI,
PEMO-Q-HI,
NCM
ModA, SRMR-HA
NON-
ENHANCED
STOI SRMR-CI ALL EXCEPT
HASPI
(PEMO-Q)
ALL (ModA)
ENHANCED STOI SRMR-CI PEMO-Q-HI,
HASQI, PEMO-
Q
ModA
NFC – – PEMO-Q-HI,
STOI
SRMR-HA
comp
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
®