The im paired-hearing group showed a trend of decreasing
%DLFc with increasing
frequency. In term s ofDLFc
(Hz), the pattern was one of an increase inDLFc
from 400 to 800 Hz, and then approxim ately constantDLFc
(Hz) across 800 and 1200 Hz. However, no significant difference in%DLFc
was found across the Fc frequency range. For the norm al-hearing listeners, an increase inDLFc
(Hz) with increasing frequeney was found for an Fc greater than 800 Hz, leading toapproxim ately constant
%DLFc.
In many studies (for exam ple, the single form ant discrim ination studies of Lyzenga and Horst (1995) and Gagné and Zurek (1988)) the increase inDLFc
(Hz) with frequency is attributed to the increased bandw idth o f auditory filters with increasing frequency. By analogy, it might be suggested that the constantDLFc
(Hz) seen in the profoundly hearing-im paired group is consistent with constant auditory filter bandw idths across the 400 to 1200 Hz frequency range. There are two lines of evidence against this view, however.Firstly, in the attempts at predicting
DLFc
for the im paired-listeners using excitation-patterns, thresholds did not com e near to m atching the observed data by m anipulations to auditory filter bandw idth alone. Increases in predictedDLFc
w ere found, but the effects were much too small to approxim ate the largeDLFc
show n by the hearing-im paired group. The lack of a significant effect was attributed to the nature of the discrim ination task, whereby level changes in the slopes of the stim uli either side of the form ant centre-frequency could easily be detected by a minimal am ount o f auditory filtering. The small effects of increased bandw idth that were found w ould be enough to account for the increasedDLFc
(Hz) with increasing frequency found for the norm al-hearing group data, but not for the hearing-im paired group.Secondly, the pure-tone discrim ination data and predictions for the im paired group both show a trend of increasing DL (Hz) with increasing frequency (group mean DL (Hz) = 23.0, 34.7, 64.1, 119.0 and 335.6 at 0.1, 0.2, 0.4, 0.8 and 1.2 kHz respectively). The statistical analysis showed that these observed DLs were not significantly different when expressed as a percentage. The pure-tone predicted DLs were shown to be affected by increases in auditory filter bandw idth, but the broadening factor was limited to 3.8 and increases to the threshold criterion were used to elevate predicted thresholds. If excitation patterns mediate pure-tone discrim ination for these listeners, these data suggest increasing auditory filter bandw idth with increasing frequency.
A model based on impaired excitati on patterns with an increased threshold
criterion se e ms to lit the obser ved data r easonabl y well. T h e e v i de n ce suggests
that residual f r equency analysis is sufficient to locate single formant peaks with a
resolution on the or de r o f 140 to 270 Hz for Fc o f 4 0 0 to 1200 I Iz. Thi s resolution
is not affected by Fo for Fos up to 200 Hz. Figure 7 s ho ws impai red excitation
patterns based on the expe r i ment al stimuli. The excitation patterns are der ived in
the s am e wa y as for the model predictions desc ri bed earlier. For each panel, the
exc it at i on pattern for the standard formant is s h ow n by the solid line. T h e dashed
line represents the standar d formant shifted by an amo un t c o r r es p on d i n g to the
o bs er v e d DLFc, and the dot ted line represents the formant shifted by the predicted DLFc. For the predicted DLFc excitation patterns, a level difference o f 7.0 dB exists bet we en the excitation patterns o f the st andar d and the just di scr i mi nabl e
different stimulus. 120- F a ( r o f ) = 1 0 0 H z : F I = 4 0 G H > — ^ - - F o ( r a f ) = lOD Hz :: F I = aOD Hz ■" F a ( p o f ) = IDO Hz :: F J = 120D H z l i a 100 00 80 m - F o tr e f ) = 200 Hz : FI = 400 Hz - - F c^ref) = 2 0 0 Hz :: FI = aoo Hz - - F o (r« f] = 200 Hz F) = 1200 Hz U D 100 9 0 80 70 0.10 1.00 2 M Frequency (kHz)
Fieure 7. Impai red excit ation patterns for stimuli in D L F r task, with Fc at
Figure 7 shows that, at threshold, the largest level differences between the standard and actual or predicted excitation patterns occurs on the (steeper) low- frequency side. It can be seen that for a broadening factor of 3.8, the excitation- pattern is able to represent both the high and low slopes o f the stim uli, even for an
Fc
of 1200 Hz, where frequeney selectivity is likely to be more im paired than for 400 or 800 Hz. If DLFc perform ance was based on excitation patterns. Figure 7 provides some insight as to why there was a laek of any effect of Fo onDLFc
means. For both Fq = 100 and 200 Hz panels, there are no cases where individual harm onics are resolved. The shapes of the standard excitation patterns are essentially identical between the two FqS. The lack of any significant difference betweenDLFc
perform ance for the two Fo conditions is consistent with the im paired excitation-pattern explanation.In summary, various lines of evidence suggest that perform ance in the
DLFc
task was mediated by an impaired excitation-pattern process in conjunction with a raised threshold criterion. A model based on normal excitation patterns was able to give good fits to the observed norm al-hearing data. It has not been dem onstrated that impaired excitation patterns are the sole m echanism used by these listeners in the discrim ination task. The role of a temporal m echanism has not been explored in this study. The need for a raised threshold criterion m ight reflect the deficit in temporal processing after auditory filtering, i.e. at the output of the auditory filter. In this sense, the raised criterion m ight be m im icking the degraded tem poral processing o f these listeners.The
DLFc
values found in this study are well in excess of the6.0
% m inim um difference in F I and F2 frequencies of British vowels cited by Rosen and Fourein (1986). Thus, normal discrim ination between vowels would not occur if perform ance were based on FI and F2 frequencies.5.3.
Perform ance inDLFn task.
It was originally hypothesised that
DLFo
discrim ination w ould be m ediated by a tem poral process. The fixed spectral envelope was intended to elim inate cues based on changes in excitation pattern, and individual harmonics were thought unlikely to be resolved except, perhaps, for the Fo,ef = 200 Flz, Fc = 400 Hz condition. The results from the excitation pattern analysis support this hypothesis. H owever, apart from the previously m entioned condition, predicted values for theDLF()
task were greatly in excess of the observedDLFoS.
This suggests that theDLF()
results m easured in this study result from a tem poral process extracting a residue pitch from a group of unresolved harmonics. However, the pure-tone DLs w ere approxim ated by excitation-pattern predictions, suggesting a spectral process. W hen the nature of the two tasks is considered, it is not surprising that excitation patterns can predict perform ance for pure-tone DLs but notDLFo
for com plex tones. In the pure-tone discrim ination, a shift in the centre frequency of the excitation pattern occurs that corresponds to a shift in the pure-tone frequency, and the resulting level changes on the low- and high-frequency sides of the excitation pattern can be used as a cue. In contrast, the com plex toneDLFo
task involves no shift in the excitation pattern centre frequency, nor any change in the low- and high-frequency skirts (except for very large shifts inFo).
Providing that individual harmonics are not resolved (and Figure 7 suggests this is the case), the hearing-im paired listeners must use the tem porally encoded repetition rate of the stim ulus for discrimination.The single
DLFo
condition where the excitation-pattern prediction matched the data needs some explanation. W hen Fo,ef = 200 Hz, the first com ponent (at the second harmonic) was at the centre frequency o f the 400 Hz formant. Higher frequency harm onics had successively low er levels than this harm onic in steps o f6
dB. In theDLFo
task, the frequency of this harmonic co-varied w ith theFo,
and so the spectral envelope could not have stayed constant. If the higher harmonics had little effect on the processing of the first com ponent, subjects were effectively perform ing a pure-tone discrim ination task for a frequency of 400 Hz. Figure 3shows that the
Fo,er = 200, Fc
= 400 HzDLFo
condition and pure-tone = 400 Hz conditions gave near identical DLs of about 17 %. For an F(),ei o f 100 Hz, the form ant is represented by harmonics on both the low and high frequency sides, and thus spectral envelope remains constant with changes in Fq.For natural vowels, where a full com plem ent o f harm onics is present, these listeners w ould have to derive pitch from unresolved harm onics in a m anner sim ilar to that for the other com plex-tone
DLFo
conditions used here. Then,Fo
differences less than about 20 % could not be encoded. TheDLFo
values found in this study suggest that a Fo sweep of 100 to 120 Hz within a natural vowel w ould not be distinguishable by these listeners. O f course, there is a possibility (not exam ined in this study) that the com ponent at the fundam ental frequency might be resolved in this situation. The pure-tone frequency DLs at 100 and 200 Hz do not suggest that perform ance would be any better if this were the case.5.4. A udiom etric factors affecting performance.
So far, the account of
DLFc
perform ance has focused on asymm etries in the slopes o f the excitation patterns. A factor which might have affected perform ance is variation in the sensation level of the stim uli with frequency. Figure 7 shows that, as frequency increases, less and less o f the excitation pattern is above threshold. At Fc = 1200 Hz, the dynamic range over which level changes in excitation can be com pared is about 10 to 15 dB. For an Fc of 400 Hz, the dynam ic range is in the range 30 to 35 dB. Thus, the sensation level of the stim uli at the two extrem es of frequency is very different. During the discrim ination task, changes in the sensation level of the signal stim ulus w ould have occurred as the centre frequency moved up. The equal loudness correction function was intended to give equal sensation levels o f com plex tones across frequency. However, the decreasing dynamic range (between threshold andMCL)
meant that com ponent levels did not always increase in proportion to the threshold as frequency increased. Thus, an increase in formant frequency could have meant a reduction in suprathreshold excitation. For the subjects in the experim ent, the level Jittersupposedly rem oved overall loudness cues. Such cues may have persisted for extrem e changes in threshold and M CL with frequency. It is difficult to rule out such cues, and it may be that DLFc for the high Fc conditions is low er than for low Fc conditions for this reason. The excitation-pattern predictions used in this study were not affected by changes in overall level, yet the predictions gave DLs som ew hat lower than the observed DLs.
5.5. Im plications of DLs for vowel perception.
The prim ary aim o f these experim ents was to reveal the frequency resolving potential of the profoundly hearing-im paired listener using stimuli that have a broad approxim ation to vowel sounds. The DLFc results im ply lim itations in the I ange o f distinguishable form ant frequencies available to this group, while the DLF() results indicate limitations to the prosodic inform ation conveyed by these vowels. A lthough the DLs are generally high and should preclude the use of subtle frequency cues in conveying vowel inform ation, there have been some encouraging outcomes. It was suspected that Fc discrim ination w ould be at its best at low frequencies, and would break down for high frequencies where frequency selectivity would be more degraded. In fact, some useful degree o f discrim ination was found across the frequency range tested. Although frequencies above 1200 Hz were not tested, it is possible that the useful range could extend beyond this. Over the range tested here, the DLFc values suggest perhaps up to four or five of these single formants could be used to signal distinguishable tim bre percepts to these listeners. If vowel perception can be approxim ated by a centre of gravity of form ants (Chistovich and Lublinskaya, 1979) or other inform ation-collapsing approach, the prospects for signalling the timbre of vowels are not so bleak for these listeners. However, if detail of spectral envelope related to both FI and F2 inform ation is required, these listeners w ould struggle somewhat. The next issue in this regard is perform ance with tw o-form ant vowels.
The F() discrim ination of unresolved harmonics found in the study shows some potential in discrim inating the pitch o f voiced speech. At a basic level, these
listeners should be able to distinguish average male from female Fo. A lthough not explicitly tested in this study, the results also suggest a potential for discrim inating between Fq sweeps which change by more than 20 %. In terms of speech
processing, the results suggest that it may be possible to exclude the Fq harmonic
of a signal from voiced speech w ithout any detrim ental effect on pitch perception. These listeners perform ed as well with a pure tone at 100 Hz as with a com plex tone com posed o f unresolved harmonics. The m issing Fqcom ponent might reduce
the upward spread of masking to low-frequency formants, and allow greater Fc resolution. In this respect, the results support ideas put forward by Fourein (1990), who argued that hearing-im paired listeners would benefit in speech perception perform ance from a reduction in the spectral com plexity o f speech sounds.