Memoire Online - Contrainte Psycho-Physiques et Electrophysiologiques sur le codage de la stimulation électrique chez les sujets porteurs d'un implant cochléaire

L'objectif de cet article est d'étudier la qualité du traitement du signal effectué par le processeur vocal Digisonic sur l'enveloppe temporelle du signal acoustique.

Nous avons comparer les pouvoirs de discrimination des asymétries temporelles chez les sujets implantés cochléaires Digisonic dans les conditions suivantes : via le processeur ou en stimulation directe.

Les résultats montrent une dégradation des pouvoirs de discrimination due au traitement du signal effectuer par le processeur.

'Laboratoire de Psychologie Expérimentale, Institut de Psychologie, Paris, France, ²Laboratoire de Physiologie Sensorielle, Hôpital E. Herriot, Lyon, France and ³Centre for the Neural Basis of Hearing, University of Cambridge, UK

This paper presents a study in which five cochlear implantees were asked to discriminate the timbre of stimuli with temporally asymmetric envelopes. Stimuli were damped and ramped sinusoids presented acoustically. They were transformed by the speech processor of the implant and were presented through one electrode. All cochlear implantees could discriminate the damped and ramped sinusoids when the half-life was 4 ms, the carrier frequency was 400 Hz, and the period of the envelope was 50 ms. In a second experiment, timbre discrimination performance was measured as a function of half-life for two cochlear implantees. Both showed that timbre discrimination was possible over the range 1-24 ms. In normalhearing listeners, the range is 1-64 ms and in cochlear implantees, stimulated directly without the speech processor, the range is 1-300 ms. At long half-lives, the decrease in discrimination performance observed with the speech processor appears to be due to the amplitude compression applied by the device. The present results suggest that it may be important to ensure that cochlear implants do not restrict temporal asymmetry unduly when applying compression to control level.

Key words: timbre perception, temporal asymmetry, cochlear implant, amplitude compression, speech processor

Two sounds with identical magnitude spectra can have very different sound quality or timbre. Among the acoustical cues used by listeners with normal hearing to identify timbre, temporal envelope cues such as onset and offset transients play an important role (for a review, see Handel, 1995). In cochlear implantees, temporal envelope cues are coded by changes in the amplitude and time pattern of stimulation on individual electrodes. Past studies with single-channel cochlear implants suggested that these changes may give rise to changes in perceived timbre but the results were not conclusive. Dobie and Dillier (1985) asked two cochlear implantees to discriminate triangular and trapezoidal waveforms

Address for correspondence: C. Lorenzi, Laboratoire de Psychologie Expérimentale, URA CNRS 316. Institut de Psychologie, Université René Descartes, Paris V. 28, Rue Serpente, 75006 Paris, France

from square waveforms. One cochlear implantee discriminated these waveforms remarkably well by labelling them `sharp' or `dull'. However, this labelling was inconsistent between different days of testing. The other cochlear implantee discriminated the triangular and square waveforms as well as the first cochlear implantee but was unable to label them.

Changes in the rate of onset (attack) and offset (decay) of a soundwave may be regarded as changes in the asymmetry of its temporal envelope. In a series of experiments performed with normal-hearing listeners, Pa tterson (1994a, b) showed how the effect of temporal asymmetry on the perception of timbre could be studied systematically using damped and ramped sinusoids. The term `damped sinusoid' referred to a segment of a sinusoid with a damped exponential envelope that was repeated cyclically to produce a sustained sound. The `ramped sinusoid' was

simply the damped sinusoid reversed in time. When the half-life of the exponential is 4 ms, normal-hearing listeners hear the damped version as a unitary source (a roll on a drummer's wood block), whereas the ramped version is heard as a co-ordinated pair of sounds (a roll on a soft leather table top accompanied by a continuous sinusoid). The effect is important because the time-reversal affects the temporal envelope of the soundwave without changing its magnitude spectrum. The results of these experiments showed that normal-hearing listeners could discriminate the timbre of a damped sinusoid from that of a ramped sinusoid when the half-life is in the range 1-50 ms.

The change in perceived timbre elicited by ramped and damped envelopes has been investigated recently in cochlear implantees (Lorenzi et al., 1997). In this experiment, ramped and damped current pulse trains were delivered directly to a single electrode of the implant without going through the pre-processor of the implant. The results showed that, when the level of the stimuli is adjusted to fit their audibility range, cochlear implantees can distinguish between ramped and damped envelopes over a much wider range (1-300 ms) than normalhearing listeners. Unlike the discrimination data of Dobie and Dillier (1985), the data of Lorenzi et al. (1997) are highly consistent across cochlear implantees. The better-than-normal performance of cochlear implantees indicates that asymmetry in the temporal envelope of a sound is a powerful cue for timbre identification in these listeners. It also suggests that, in normal-hearing listeners, cochlear compression limits the sensitivity to temporal asymmetry. The speech processor used in most implants includes compression intended to simulate the compression applied by the intact cochlea. This suggests that the compression in the speech processor degrades the perception of temporal asymmetry in implantees as it does in normalhearing listeners.

Five cochlear implantees were asked to discriminate the timbre of ramped and damped sinusoids when presented acoustically through the speech processor of a Digisonic DX10 cochlear implant. The results of this experiment were compared with those obtained previously by direct electrical stimulation (Lorenzi et al., 1997).

Five post-lingually profoundly deaf listeners (BM, DL, LR, RF, SP) participated in these experiments, three of whom (BM, LR and SP) also participated in the experiments reported by Lorenzi et al. (1997). They were all experienced in twointerval, two-alternative forced choice (2I, 2AFC) tasks. Clinical information about these patients is presented in Table 1. Their audiometric thresholds at 0.5, 1, 2 and 4 kHz are presented in Table 2. They were all implanted with a Digisonic DX10 device (MXM), which is a transcutaneous 15channel cochlear implant with an intracochlear electrode array (Beliaeff et al., 1994). Stimuli were presented acoustically to cochlear implantees and were transformed through the speech processor of the Digisonic DX10 cochlear implant. The device performs a 128-point Fast Fourier Transform (FFT) from 100 to 7800 Hz. The device imposes an absolute threshold of 40 dB SPL and it applies logarithmic compression above this threshold, separately in each channel. The compression device does not include any dynamic elements (e.g. AGC attack and decay times). The volume control of the processor was fixed during the testing period. Activation was limited to the most apical electrode which delivered monophasic (capacitively coupled) current pulses; the remaining 14 electrodes in the array were connected together to serve as the return path for the current. More specifically, the return path was a mixture of 'corn- mon ground' and monopolar modes of stimulation. Radiography revealed that electrode positioning was roughly the same for all implantees. The activated electrode was assigned a single wide frequency band (100-7800 Hz), which combined the 64 energy values produced by the 128-point FFT. For each patient, the pulse duration (in lis) was adjusted from threshold (MM) to comfort level (Max). The Min and Max values for each cochlear implantee are presented in Table 1. The carrier was a train of monophasic (capacitively coupled) current pulses with a rate of 400 pulses per second which was the maximum pulse rate provided by the speech processor. For comparison, three listeners with normal audiometric thresholds also participated. As implantees, they were highly experienced in 21, 2AFC tasks.

Patient	Age (years)	Cause of deafness	Duration of implant use (months)	Min (ps)	Max (ps)
Br '	64	Head trauma	7	12	55
DL	30	Progressive deafness	15	86	110
LR	60	Unknown	3	10	45
RF	69	Progressive deafness	9	10	41
SP	44	Head trauma	6	14.5	45

Table 2. Unaided air-conduction thresholds in dB HL at the left (L) and right (R) ears of the five impaired listeners of the study

	L	0.5 R	L	1 R	L	2 R	L	4 R
Patient
BM	115	115	>120	>120	>120	>120	>120	>120
DL	105	105	100	120	90	>120	95	>120
LR	115	105	>120	115	>120	>120	>120	>120
RF	110	105	115	>120	>120	>120	>120	>120
SP	105	90	100	100	90	100	95	115

where fis the carrier frequency (400 Hz), Ais the starting amplitude, hl is the half-life of the damped sinusoid, and c is a constant (-0.693147), that brings the envelope to A/2 in hl T is the repetition period which is 50 in both experiments. The ramped sinusoids were produced by reversing the damped sinusoids in time. The stimuli were digitally generated by a 16-bit D/A converter at a sampling frequency of 44.1 kHz. The duration of the stimuli was 500 its; the silent inter- val between stimuli was 500 Fts. The stimuli were presented in free field through a loudspeaker positioned at 0° azimuth and 0° elevation. Listeners sat at 1 m from the loudspeaker, and were asked to face it during the course of the experiment. The loudspeaker was a full range (150 Hz to 20 kHz) driver. The stimuli were presented at a moderately loud level with the damped and ramped sinusoids having the longest half-life set to 65 dB SPL (SPL was measured with a sound level meter placed at the listener's head position). As the half-life decreases, the energy and the loudness of the sound decreases. To maintain the stimuli at the same loudness, the maximum

amplitude, A, was increased by the square root of 2 each time the half-life was decreased by a factor of 2.

Segments of the acoustic stimuli are presented in the upper panels of each section of Fig. 1; the left and right columns show damped and ramped stimuli, respectively. The half-life of the exponentiel is 1 its in the top section, 8 ps in the middle section, and 64 Fis in the bottom section. The trains of stimulation pulses produced by the activated electrode in response to each sound wave were recorded by a specially designed computer interface (Digigram system, MXM). They are presented in the panels below each sound wave. In these panels, the ordinate is in microseconds because the amplitude of the acoustic stimulus is coded by the duration of the current pulses produced by the Digisonic DX10 device. The trains of current pulses show that the speech processor of the implant degrades the temporal envelope of the acoustic stimulus. When the halflife is 1, 8, or 64 ils, the half-life of the electric waveform is longer than that of the acoustic waveform. When the half-life is 1 ps or 64 ils, the envelope asymmetry is largely lost. However, asymmetry is preserved when the half-life is 8 ils.

0.5
0.0

· -0.5

-1.0

e 120

100 -

4
· 80

-0 60 - g)

· 40 - a.

Ift

Fig. 1. Segments of the acoustic stimuli are presented in the upper panels of each section. The left and right columns show damped and ramped sinusoidal waves, respectively. The half-life of the exponential is 1 ps in the top section, 8 ,us in the middle section, and 64 ps in the bottom section. The carrier frequency is 400 Hz. The repetition period is 50 ,us. The train of current pulses produced by the activated electrode in response to each sound wave is presented in the panel below each sound wave. In these panels, the ordinate (the pulse duration) is in microseconds.

Fig. 2. Performance of the five cochlear implantees when the half-life is 4 ,us. Stimuli were presented acoustically and went through the speech processor. Two tasks were used to mea- sure discrimination performance. In the first task (unfilled bars), implantees were asked to say if the two sounds were identical or different. In the second task (filled bars), implantees had to choose the interval containing the sound with the more `drum-like' quality. Performance corresponds to the percentage of damped sounds as having the more drum-like quality.

cochlear implantees stimulated directly without the use of the preprocessor (Lorenzi et al., 1997).

In the second experiment, psychometric functions were measured for cochlear implantees BM and DL using procedure B (`more drum-like sound quality' task). The half-life was systematically varied from 1 to 32 ms. The psychometric functions of BM and DL (solid lines with open and filled circles, respectively) are presented in Fig. 3. In both cases, the shortest just-discriminable half-life was between 1 and 1.5 and the longest just-discriminable half-life was between 16 and 24 us. At shorter and longer half-lives, discrimination performance fell off abruptly. For comparison, psychometric functions were measured for three normal-hearing listeners with half-life varying from 0.125 to 128 p. The dotted line with open triangles shows their mean performance. The dashed line without symbols shows the mean of the data obtained by Patterson (1994b) and Irino and Patterson (1996) in similar

In the first experiment, two types of discrimination procedure were used, both of which were 21, 2AFC. In procedure A, listeners were presented a ramped or a damped sinusoid chosen at random in each interval. Thus, there were four possible pairs of sounds: ramped/ramped, damped/ damped, ramped/damped and damped/ramped. They were asked to say if the two sounds were identical or different. In procedure B, they were presented a damped sinusoid in one interval and a ramped sinusoid with the same half-life in the other interval, and asked to choose the interval with the 'more drum-like sound quality'. Thus, performance corresponds to the percentage of damped sinusoids chosen as having the more drum-like quality. Procedure B was also used throughout the second experiment. In the first experiment, the half-life was fixed at 4 ps. In the second experiment, the half-life was fixed within a block and was varied from 1 lus to 32 us from block to block. Each block contained 50 trials in the first experiment and 30 trials in the second experiment.

In both experiments, listeners received visual feedback concerning the accuracy of their response after each trial. Listeners sat in a double-walled soundproof booth, in front of a keyboard connected to the computer controlling the experiment. They received 15 min of preliminary training before participating in each experiment.

The performance of the five cochlear implantees for the first experiment is presented in Fig. 2; the half-life was fixed at 4 its. Each bar is based on 50 trials; unfilled bars for procedure A (`same/different' task) and filled bars for procedure B (`more drum-like sound quality' task). For both tasks, performance was invariably well above chance: the mean performance of the five cochlear implantees was 94.4% (SD 7.12%) for procedure A, and 94.8% (SD 10.54%) for procedure B (p < 0.05 for 61% correct responses). They all heard the difference between damped and ramped sinusoids and chose the damped sinusoid as having a stronger drum-like quality without difficulty. Informal testing with cochlear implantees revealed that the ramped sounds produced the stronger tonal quality, as is the case for normalhearing listeners (Patterson, 1994a, b), and for

Fig. 3. Psychometric functions for cochlear implantees BM (solid lines with open circles) and DL (solid fines with filled circles), showing discrimination performance as a function of half-life. Stimuli were presented acoustically and went through the speech processor. Implantees had to choose the interval containing the sound with the more drum-like' quality. The data from cochlear implantees are plotted along with (1) the mean of the data obtained with three normal-hearing listeners in identical conditions (dotted line with open triangles), and (2) the mean of the data obtained with normal-hearing listeners in similar conditions by Patterson (1994b) and Irino and Patterson (1996) (dashed line without symbols).

conditions.¹ For all normal-hearing listeners, discrimination performance was above 90% when the half-life was between 1 ms and 10 ps, but at chance when the half-life was either below 0.125 or above 64 its. In summary, the results of both experiments show that cochlear implantees receiving stimuli through the Digisonic DX10 speech processor can discriminate the timbre of ramped and damped sinusoids as well as normalhearing listeners (when the half-life is 4 its, for instance). However, the results of the second experiment indicate that discrimination is restricted to a narrower range of half-lives in cochlear implantees.

'When the half-life was below 1 ps, the normal-hearing listeners of the present study showed better performance than that reported by Patterson (1994b). This better performance may be due to the fact that Patterson (1994b) did not pro- vide feedback after each response, and he mixed experimental conditions within blocks of trials. It remains, however, unclear why this methodological difference should affect discrimination performance under 1 ps, as opposed to performance over 10 ps.

The average psychometric functions for the cochlear implantees and the normal-hearing listeners who participated in the second experiment are plotted in Fig. 4 (solid Fines with filled circles for cochlear implantees, dotted lines without symbols for normal-hearing listeners). The solid lines with open circles show the mean of the data obtained by Lorenzi et al. (1997) with cochlear implantees stimulated without the intervention of the speech processor. Fig. 4 shows that bypassing the speech processor improves timbre discrimination performance; the effect being stronger at long half-lives than at short half-lives: the longest justdiscriminable half-life is increased by a factor of 25 in the direct stimulation mode. The experiments without the speech processor were performed after those with the speech processor. It is, however, unlikely that practice effects explain the better performance obtained by direct electrical stimulation as the two sets of experiments were separated by a period of 11 months. In addition, all implantees were highly skilled in 21, 2AFC tasks before participating in both experiments.

Fig. 4. Average psychometric functions for the two cochlear implantees stimulated with the intervention of the speech processor (solid fines with filled circles) and the three normal-hearing listeners who participated in the second experiment of this study (dotted fines without symbols). The data are plotted along with the mean of the data obtained by Lorenzi et al. (1997) with cochlear implantees stimulated without the intervention of the speech processor (solid fines with open circles).

The data show that cochlear implantees using the speech processor can consistently label and discriminate the timbre of ramped and damped envelopes, but over a narrower range of half-lives than implantees stimulated directly with the use the speech processor. These data demonstrate that the speech processor of the Digisonic DX10 device degrades the envelope information significantly and affects the timbre of damped and ramped sounds. The loss of fidelity shown in Fig. 1 by the current pulse functions for a half-life of 1 us and 64 us suggests that performance in cochlear implantees is limited by two components of the coding scheme of the Digisonic DX10. At short half-lives (e.g. 1 us), the persistence of the cochlear implant signal and the loss of temporal asymmetry are mainly caused by the long duration of the temporal window (8.2 ils) used for the computation of the 128-point FFT. At long half-lives (e.g. 64 us), the loss of temporal asymmetry appears to be mainly due to the compression circuitry of the device.

This paper describes cochlear implantees' ability
to discriminate and label the timbre of sinusoids

with asymmetric temporal envelopes. The stimuli were presented acoustically and were transformed by the compressive speech processor of the implant. Stimulation was restricted to a single electrode of the implant.

When the level of the stimuli is adjusted to fit their audibility range, the implantees are as sensitive to temporal asymmetry as normal-hearing listeners, but over a narrower range. In other words, timbre differences elicited by changes in the temporal envelope asymmetry of sounds are less salient in cochlear implantees using their speech processor than in normal-hearing listeners. A comparison between the discrimination performance of cochlear implantees stimulated with and without the use of the pre-processor suggests that the poorer performance of cochlear implantees stimulated acoustically is mainly caused by the 128-point FFI' at short half-lives, and by the compression circuitry of the device at long half-lives.

Temporal asymmetry is a prominent property of speech sounds and sounds produced by musical instruments. It is known to play a role in speech discrimination and timbre perception. For instance, much of the information about the nature of consonants is contained in the 10-40 following an

onset or preceding an offset, and some of the information is in the abruptness of the onset or offset (for a review, see Stevens and House, 1972). Hearing-impaired listeners are therefore likely to make good use of temporal asymmetry when it is available. The present results indicate that it may be important to ensure that cochlear implants (and hearing aids) do not restrict temporal asymmetry unduly when applying compression to control level. Compression is essential for fitting the wide dynamic range of everyday sounds (around 60 dB) into the limited dynamic range available for electrical stimulation (less than 10 dB), and the current results should not be interpreted to suggest that compression should be removed from the speech processors. Rather, the results emphasize the importance of the characteristics of the compression to use, that is, the compression function and the attack and decay times in the case of automatic gain control. In order to preclude the use of spectral cues, the cochlear implants were used in a singlechannel mode during the course of these experiments. This methodological precaution restricts the generality of the current finding. Our results will therefore need to be extended to multi-channel devices before being applied in speech processor design.

The first and third authors were at the MRC Applied Psychology Unit (Cambridge, UK) when this research was performed. The first author was supported by a post-doctoral grant from the FYSSEN Fundation. The second author was supported by a CIFRE doctoral grant from

MXM Company. We thank Prof Stuart Gatehouse , Patrick Howell, Catherine Lever and three anonymous reviewers for comments on a previous version of this manuscript.

Beliaeff M, Dubus P, Leveau J M, Repetto J C, Vincent P. Sound processing and stimulation coding of DIGISONIC DX10 15-channel cochlear implant. In: Hochmair ES, ed. Advances in Cochlear Implant. 1994; 198-203.

Dobie RA, Dillier N. Some aspects of temporal coding for single-channel electrical stimulation of the cochlea. Hear Res 1985; 18: 41-55.

Handel S. Timbre perception and auditory object identification. In: Moore BCJ, ed. Hearing. New York: Academic Press, 1995; 425-61.

Irino T, Patterson RD. Temporal asymmetry in auditory perception and a 'delta-gamma' theory of asymmetric intensity enhancement in the peripheral auditory system. J Acoust Soc Am 1996; 99: 2316-31.

Lorenzi C, Gallego S, Patterson RD. Discrimination of temporal asymmetry in cochlear implantees. J Acoust Soc Am 1997; 102: 482-85.

Patterson RD. The sound of a sinusoid: Spectral
models. J Acoust Soc Am 1994a; 96: 1409-18.

Patterson RD. The sound of a sinusoid: Timeinterval models. J Acoust Soc Am 1994b; 96: 1419-28.

Stevens K, House AS. Speech perception. In: Tobias J, ed. Foundations of modem auditory theory. New York: Academic Press, 1972; 1-62.

Lorsque l'on évalue les contraintes psycho-physiques de chaque patient, on s'aperçoit qu'il y a de grandes disparités dans la capacité à coder l'information auditive dans ses trois dimensions le temps, la fréquence et l'amplitude.

Il paraît donc important de connaître les spécificités psycho-physiques de chaque sujet implanté pour adapter au mieux l'interface bio-électrique lors du réglage.

- L'estimation de la fonction de tonie en fonction de l'électrode stimulée nous permet d'attribuer le plus fidèlement possible la répartition fréquentielle.

- L'estimation de la fonction de tonie en fonction de la fréquence de stimulation nous permet d'utiliser la plage de fréquence de stimulation adéquate pour coder le fondamental laryngé.

- L'estimation des fonctions de sonie entre chaque canal nous permet d'adapter l'énergie acoustique à chacune des électrodes.

- La mesure de la résolution temporelle va permettre de régler la fréquence de stimulation moyenne.

La mesure de chacun de ces paramètres peut facilement s'intégrer dans des protocoles rapides lors du réglage, mais ne peut être effectuée que par des sujets implantés adultes et très bien conditionnés. L'élaboration de ces tests psycho-physiques pour une population implantée cochléaire pédiatrique semble être difficile et trop long. La possibilité d'utiliser des méthodes objectives comme l'électrophysiologie pour estimer les contraintes psycho-physiques semble être prometteuse.

Contrainte Psycho-Physiques et Electrophysiologiques sur le codage de la stimulation électrique chez les sujets porteurs d'un implant cochléaire

Article 12 :

AMPLITUDE COMPRESSION IN COCHLEAR IMPLANTEES ARTIFICIALLY RESTRICTS THE
PERCEPTION OF TEMPORAL ASYMETRY

Contrainte Psycho-Physiques et Electrophysiologiques sur le codage de la stimulation électrique chez les sujets porteurs d'un implant cochléaire

Article 12 :

AMPLITUDE COMPRESSION IN COCHLEAR IMPLANTEES ARTIFICIALLY RESTRICTS THEPERCEPTION OF TEMPORAL ASYMETRY

AMPLITUDE COMPRESSION IN COCHLEAR IMPLANTEES ARTIFICIALLY RESTRICTS THE
PERCEPTION OF TEMPORAL ASYMETRY