abnormal speech processing for hearing-impaired listeners...
TRANSCRIPT
Abnormal Speech Processing for Hearing-Impaired Listeners in Frequency Regions where Absolute Thresholds are Normal
Agnès C. Léger1*, Brian C.J. Moore2, Dan Gnansia3, Christian Lorenzi1
1Ecole Normale Supérieure, Paris, France; Université Paris Descartes, Paris, France; CNRS, France, 2Dept. of Experimental Psychology, University of Cambridge, Cambridge, UK, 3Neurelec, France.
Speech intelligibility is reduced for listeners with sensorineural hearing loss. This reduction is larger for speech in noise, especially when the noise is modulated. The extent to which this reduction is due to reduced audibility or to supra-thres-hold de�cits is still debated.The hypothesis that supra-threshold de�cits have a speci�c in�uence on speech intelligibility was tested. The e�ect of audi-bility was controlled by measuring speech intelligibility for hearing-impaired listeners using nonsense speech signals �lte-red in low- and mid-frequency regions, where pure-tone sensitivity was normal or near-normal.
MethodSpeech identi�cation test (consonant identi�cation) using Vowel-Consonant-Vowel (VCVs) signals presented in quiet and in several noise maskers. Noise maskers were either unmodulated, or modulated in amplitude or in frequency.Speech and noise maskers were �ltered into various frequency regions: - low-frequency region (<1.5 kHz) - mid-frequency region (1-3 kHz) - low+mid frequency region (<3 khz)
Results
0
10
20
30
40
50
60
70 Normal-hearing listeners
Hearing-impaired listeners
amplitude modulated, long gapsamplitude modulated, short gapsspectrally modulated, large gaps
spectrally modulated, small gapsspectrally modulated, medium gaps
Quiet
unmodulated
−6 −3 0 quiet0
10
20
30
40
50
60
70
SNR (dB)
Normal-hearing listeners
−6 −3 0 quietSNR (dB)
Hearing-impaired listeners
Noise maskers:
Low
-freq
uenc
y re
gion
(<1.
5 kh
z)M
id-fr
eque
ncy
regi
on (1
-3 k
hz)
Mea
n sp
eech
iden
ti�ca
tion
scor
e (R
AU
)
Listener
1011
825
414
19 3
122
9 2
724
23 6
518
2620
1317
1216
1521Mean
Low
+Mid
(≤3
kHz)
freq
uenc
y re
gion
Indi
vidu
al s
peec
h id
enti�
catio
n sc
ore
(RA
U)
Low
(≤1.
5 kH
z) fr
eque
ncy
regi
onM
id (1
-3 k
Hz)
freq
uenc
y re
gion
0
25
50
75
100
0
25
50
75
0
25
50
75
QuietNoise
Control group
Good performersPoor performers
Listeners
ConclusionsHearing-impaired listeners with hearing loss in high-frequency regions showed mild (Léger et al., 2012a) to severe (Léger et al., 2012c) intelligibility de�cits for speech both in quiet and in noise in these frequency regions of near normal audibility. Similar de�cits were obtained for speech in unmodulated and modulated masking noises (Léger et al., 2012a). These results provides additional evidence that speech intelligibility may be strongly in�uenced by supra-threshold auditory de�cits, and suggest that audiometric thresholds within the “normal” range (better than 20 dB HL) do not imply normal auditory function.
Introduction
Do supra-threshold de�cits have a speci�c in�uence on speech intelligibility?
What is the nature of these supra-threshold de�cits?Reduced audibility? Maybe – Signi�cant correlation between speech identi�cation scores and pure tone average in the tested frequency region for Léger et al. (2012c), but not for Léger et al. (2012a). – Predictions of the scores based on audibility (ESII) do not signi�cantly correlate with the obtained scores (Léger et al., 2012a; 2012c).
Reduced spectral resolution? Yes, but not entirely – Correlation between speech identi�cation scores and transient oto-acoustic emissions (OAEs) and OAE signal-to-noise ratios averaged in a mid-frequency region (Léger et al., 2012c).
0
20
40
60
80
0
20
40
60
0
Mea
n id
enti�
catio
n sc
ore
(RA
U)
Low
-freq
uenc
y re
gion
(≤1.
5 kH
z)
1 1.3 1.7 30
20
40
60
0
1 1.3 1.7 3Smearing factor
1 1.3 1.7 3
0
20
40
60
80
0
20
40
60
0
Mea
n id
enti�
catio
n sc
ore
(RA
U)
Mid
-freq
uenc
y re
gion
(1-3
kH
z)
1 1.3 1.7 30
20
40
60
0
1 1.3 1.7 3Smearing factor
1 1.3 1.7 3
amplitude modulatedlong gaps
unmodulated
amplitude modulatedshort gaps
spectrally modulatedlarge gaps
spectrally modulatedmedium gaps
spectrally modulatedsmall gaps
quiet
amplitude modulatedlong gaps
unmodulated
amplitude modulatedshort gaps
spectrally modulatedlarge gaps
spectrally modulatedmedium gaps
spectrally modulatedsmall gaps
quiet
NH listenersLéger et al. (2012)
HI listenersLéger et al. (2012)
– Simulations (using spectral smearing, Baer & Moore, 1993) for normal-hearing listeners of the average or maximal broa-dening of the auditory �lters expected for the hearing-impaired listeners leads to a degradation of speech identi�cation scores (Léger et al., 2012b). However, it does not account entirely for the de�cits demonstrated by the hearing-impaired lis-teners.
ListenersNormal-hearing listeners, and hearing impaired listeners with a hearing loss in the high-frequency region but normal (≤20 dB HL) or near-normal (≤30 dB HL) audiometric tresholds in the tested frequency regions.
Abnormal temporal enveloppe processing? No (?) – Decreasing the duration of the temporal gaps in the amplitude-modulated masker has a comparable (deleterious) e�ect on speech intelligibility for normal-hearing and hearing-impaired listeners (Léger et al., 2012a).
Reduced sensitivity to temporal �ne structure? Not in quiet, maybe in noise – No correlation between binaural estimates of temporal �ne structure sensitivity (TFS-LF test; Hopkins and Moore, 2010) and speech identi�cation scores in quiet (Léger et al., 2012c).
E�ect of age? Maybe – Léger et al. (2012a): No di�erence in identi�cation scores between two age groups for hearing-impaired listeners (41 and 72 years old); and no correlation between age and identi�cation scores for hearing-impaired listeners. – Léger et al. (2012c): Signi�cant correlation between age and identi�cation scores for normal-hearing and hearing-im-paired listeners.
Acknowledgments & References*A.C. Léger is now at Research Laboratory of Electronics, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA. Email address: [email protected] work was supported by a grant from the Royal Society (International Joint Project, 2009R3). A.C. Léger and C. Lorenzi were supported by 7FP-SME222291 Dual-Pro program. A.C. Léger was supported by a CIFRE grant from ANRT and Neurelec. B.C.J. Moore was supported by the MRC (UK, grant number G0701870). C. Lorenzi was supported by two grants (HEARFIN and PRESBYCUSIS Projects) from ANR.
Baer, T., & Moore, B. C. J. (1993). E�ects of spectral smearing on the intelligibility of sentences in noise. The Journal of the Acoustical Society of America, 94, 1229–1241. Hopkins, K., & Moore, B. C. J. (2010). Development of a fast method for measuring sensitivity to temporal �ne structure information at low frequencies. International Journal of Audiology, 49(12), 940–946. Léger, A. C., Moore, B. C. J., Gnansia, D., & Lorenzi, C. (2012b). E�ect of spectral smearing on temporal and spectral masking release in low- and mid-frequency regions. The Journal of the Acoustical Society of America, 131(5), 4114–4123. Léger, A. C., Moore, B. C. J., & Lorenzi, C. (2012c). Abnormal speech processing in frequency regions where absolute thresholds are normal for listeners with high-frequency hearing loss. Hearing Research, , 294, 95-103. Léger, A. C., Moore, B. C. J., & Lorenzi, C. (2012a). Temporal and spectral masking release in low- and mid-frequency regions for normal-hearing and hearing-impaired listeners. The Journal of the Acoustical Society of America, 131(2), 1502–1514.
Léger et al. (2012a)
Léger et al. (2012a) Léger et al. (2012c)
0
20
40
60
80
100
Aud
iom
etric
thre
shol
d (d
B H
L)
NHHI
0.25 0.5 1 2 4 8Frequency (kHz)
0.25 0.5 1 2 4 8Frequency (kHz)
Low-frequency region (<1.5 kHz)
Mid-frequency region (1-3 kHz)
0
20
40
60
80
100
Aud
iom
etric
thre
shol
d (d
B H
L)
0.25 0.5 1 2 4 8Frequency (kHz)
Léger et al. (2012c)
HI listeners show mild de�cits in all conditions compared with NH listeners. Their de�cits are larger in noise than in quiet, but are similar accross noise conditions.
A subgroup of HI listeners (poor performers) show severe de�cits in
quiet and in noise compared with older HI and NH listeners (good
performers) and young NH listeners (control group).