FORMACIÓN AUTONOMA

To understand consonant recognition in CI users, it is first necessary to be able to describe how it differs from NH listeners. By “how” is meant “what pattern of feature error pattern?” Hence two questions can be addressed:

• How does consonant recognition differ between CI users and normal-hearing listeners?

• What effect does noise have on consonant recognition in CI users?

Additionally, the relative importance of spectral and temporal resolution was raised in relation to CI users’ consonant recognition: Hence:

• Are deficits in consonant recognition in CI users due primarily to interference with temporal processing, with spectral processing, or equally with both?

More broadly, it is not adequate to simply characterise abnormalities in CI users’ consonant recognition in quiet and noise without then defining:

• What factors affect consonant recognition in CI users?

With regard to the first question, the relevant literature shows that, first of all, consonant recognition in CI users is markedly worse than in NH listeners (even in

in consonants in CI users is significantly worse than manner or voicing perception, to a far greater extent than is the case with NH listeners. Every study that has evaluated CI users’ consonant feature information transmission has found that perception of place of articulation is poorer than manner and voicing in quiet (Donaldson and Nelson, 2000; Dorman et al., 1991; Van Tasell et al., 1992; Dorman et al., 1990; Dorman et al., 1991; Dorman, 1995). Figure 2.3 shows performance across a number of studies for voicing, place and manner in quiet. (Studies included in the chart are restricted to those studies in which CI users’ consonant recognition abilities were tested and analysed in terms of information transmission of consonant features.) The figure also includes data collated by the author for over 60 adult CI users from the South of England Cochlear Implant Centre (SOECIC). Two further details of these studies should be noted: first, all the studies used performance in quiet and, second, with the exception of Geurts and Wouters (1999), all studies undertook consonant recognition measures using VCV nonsense syllables of the form /aCa/, e.g. where the vowel /a/ precedes and follows the target consonant. Data are presented for the “best” performance conditions for those studies where comparisons of different listening or processing parameter conditions were undertaken.

0 10 20 30 40 50 60 70 80 90 100 Fu and Shannon 2000 Loizou and Poroy 2000 Geurts and Wouters 1999 Munson et al 2003, better performers Munson et al 2003, w orse performers Verschuur 2005 SOECIC % i n fo rm ati o n tr an sm is s io n voicing place manner

Figure 2.3. Consonant voicing, place and manner transmission from studies of CI user performance

The accompanying table 2.2 shows further details of the studies, although the table is not fully comprehensive in terms of the many variations across studies, which differed in other aspects of consonant confusion analysis, e.g. choice and number of stimuli, number of repetitions per stimulus, male vs. female vs. mixed speaker and number of tokens per speakers (although it is important to note almost all studies cited used consonant recognition in the /aCa/ vowel environment). Moreover, studies varied by subject parameters, e.g. CI devices, processing strategies and baseline speech

perception abilities. Given the heterogeneity of both stimulus and subject

characteristics across studies, it is interesting that the “worse place performance” pattern of results is so consistent. Although absolute levels of transmission vary between the studies, relative transmission across features is less variable. Moreover, it should be noted that NH listeners show transmission levels approximating 100% for equivalent stimuli in quiet and therefore none of the features can be said to be transmitted “normally”, at least when averaged across a group of CI users.

Table 2.2. Parameters for data sets in figure 2.3.(Further details of implant types are given in 2.3)

Study Parameters Method Implant

Fu and Shannon 2000

500 pps/ch x 4 CIS

16AFC,aCa, 2 tokens x 2 reps,

mixed gender N22

Loizou and

Poroy 2000 2100 pps/ch x 6 CIS 20 AFC, averaged across aCa, iCi, uCu, female MED-EL Geurts and

Wouters 1999 CIS averaged across aCa, iCi, uCu,initial consonants, mixed gender LAURA Munson et al

2003, better

performers mixed 19AFC, aCa, mixed gender N22, Clarion

Munson et al 2003, worse

performers mixed 19AFC, aCa, mixed gender N22, Clarion

Verschuur 2005 >1500pps/ch, 12 channels 20AFC, aCa, female MED-EL

SOECIC mixed 20 AFC, aCa, female N24, N22, MED-El

Several authors have suggested that this discrepancy between place and manner/voicing perception can be explained by the fact that CI users' spectral resolution is relatively poor compared to that of NH listeners (Dawson et al., 2000; Dorman and Loizou, 1997; Dorman et al., 2000; Loizou et al., 1999; Loizou et al., 2000b), whereas temporal processing is less impaired when compared to NH

listeners’ abilities (Busby et al.; 1993;Hescot et al.; 2000;Shannon, 1992). However, a distinction should be made between underlying psychophysical capacity as against

information loss associated with CI processing. Both CI processing and

electrical/neural interface factors may be implicated in poor place of articulation transmission. Spectral resolution is reduced by the way in which the CI transforms the signal into a relatively small number of envelope values (up to 22 depending on device), but spectral information could also be further affected by spread of excitation in the electrical/neural interface (these factors are discussed in 2.3.2 and 2.4.1).

Although the research literature has emphasised poor place performance, it is still worth noting that voicing and manner are still poor compared to normal performance. NH listeners obtain near to 100% in the listening conditions of the tests (e.g. in quiet at 60 dB SPL or greater). Therefore, if manner and voicing do rely on temporal envelope information then it follows that temporal envelope perception must also be impaired in CI users compared to NH listeners, whether because of information loss due to CI processing, the electrical/neural interface or the central auditory nervous system.