Temporal coding

It is possible that the electrical/neural interface may introduce loss of temporal information as well as loss of spectral information. Section 2.4.3 implied that TMTFs were uniform across CI users; however, a number of earlier studies showed that there was considerable variation in TMTFs between individual CI users, and in one study, this was shown to be correlated with consonant recognition, suggesting that temporal aspects of electrical/neural interface information loss may be as or more important than spectral aspects in determining individual variations in consonant recognition.

Busby et al. (1993) measured perception of temporal modulations in a group of adult CI users. They found that the shape of the TMTF also approximated a low pass filter with a cut-off frequency between 50 and 100 Hz, slightly lower than was the case for the Shannon (1992) study. What is interesting in the context of a discussion of the electrical/neural interface is that Busby et al. (1993) attempted to match temporal processing characteristics with patient characteristics, in particular duration of deafness. They found that four postlingually deafened subjects were better able to perceive temporal information than three prelingually deafened subjects. It is not possible to determine whether these variations were to do with neural or central function, but they do suggest that temporal processing varies across CI users.

Other studies have also looked at the relationship between CI users’ basic temporal psychophysical abilities and the level of speech perception they obtain. Cazals et al. (1991) measured perception of a silent gap in noise and interval between two clicks in five users of the Ineraid CI. They found that there was a relationship between

perception of click interval at the most basal CI used and perception of consonant place of articulation. The most striking evidence of such a relationship is given by Fu (2002), who found a strong correlation between consonant recognition scores and mean modulation detection thresholds across users’ electrical dynamic range. Subjects were nine users of the Nucleus 22 CI system using the SPEAK speech processing strategy. Whereas previous studies had linked speech perception abilities to TMTF performance at high input levels, Fu (2002) measured the TMTF across a range of stimulus levels and found that the mean score averaged across input levels was a significant predictor of both consonant and vowel intelligibility.

In order to convey temporal information to the CI user, the neural discharge pattern in response to CI stimulation must convey the temporal detail in the input signal. An important difference in temporal coding between acoustic and electrical hearing lies in the stochastic relationship between acoustic input and the response of the auditory nerve to stimulation. This enables high rates of temporal coding in the auditory system, up to around 4 kHz, because of the summation of neural responses across neural populations, rates which cannot be supported by individual neurons. Without stochastic resonance, phase-locking of individual nerve fibres would prevent coding

mechanism of stochastic resonance is thought to be cochlear in origin, it can be presumed that this does not occur with CI systems. Therefore, electrical hearing may be at some disadvantage with respect to coding of high frequency temporal

information. A number of papers using physiological outcome measures have

suggested that the use of very high rates and also the use of conditioning noise stimuli may improve the temporal representation and accuracy within the auditory nerve, e.g. Matsuoka et al. (2001).

There is also a question as to whether higher stimulation rates may lead to greater increased channel interaction. Brill et al. (1997) found individual variations in trade- off between channel number and stimulation rate in a group of users of the MED-EL device. It seems plausible that individual differences in this trade-off may be

mediated by the degree and nature of channel interaction. McKay et al. (2005) found that sensitivity to spectral shape was less at higher rates, given a particular number of channels. Their explanation for this was that forward masking of one pulse over a successive pulse serves to blur between-channel amplitude differences. This may help to explain why there is so much individual difference in benefits with higher stimulation rate: it is possible that individual CI users who have greater channel interaction could experience increased forward masking at higher stimulation rates compared to those with lower channel interaction.

Despite these considerations, it is appears that the focus in the present study should be on information loss associated with CI processing rather than the electrical/neural interface. It appears from the evidence presented in 2.4.2 that Nucleus 24 processing preserves temporal modulations with decreasing accuracy as modulation rates increase. Moreover, it also appeared that differences in TMTF with stimulation rate were small. Consequently, it can be hypothesised that Nucleus 24 users have little access to mid-frequency modulation frequencies (those denoting periodicity according to Rosen (1992)), no access to higher frequency modulations and that stimulation rate should make only very small differences to consonant recognition.

• Cross-channel spread of excitation has been measured in CI users using various techniques.

• Channel interaction has both a spectral and temporal aspect, although Chatterjee and Oba (2004) showed that spectral channel interaction has a stronger implication for speech perception outcomes in CI users.

• Variations in channel interaction could help to explain variations in CI user performance but the evidence base for this is limited.

• Electrode insertion is associated with an upward frequency transposition because of the alignment of the electrode array in relation to the remaining auditory elements.

• Although there is evidence that partial insertion limits performance, it is not thought that a normal insertion depth (e.g. more than 22 mm) is an important factor in limiting consonant recognition.

• There is evidence that CI users show abnormal temporal resolution, particularly at lower intensities, although the account in 2.4 suggests that some of this must be due to information loss from CI processing rather than the electrical/neural interface.

• There is also evidence of a link between temporal processing abilities and overall consonant recognition.