6 Referencias bibliográficas
FLEXIBILIZACIÓN CURRICULAR POR ESTUDIANTE Estudiante
The focus in this thesis is on the relative contribution of CI signal processing
information loss as against electrical/neural interface information loss in determining the pattern of consonant feature transmission in CI users. The overriding question motivating the research was: “to what extent can deficits in consonant recognition by CI users be explained by information loss in CI signal processing as opposed to information loss at the electrical/neural interface?” This question is not directly answerable but must be translated into experimentally tractable hypotheses. The main problem in determining the relative contribution of electrical/neural interface factors such as channel interaction is the difficulty in controlling variations in these factors between individual CI users. While there is good evidence that individual CI users may vary in terms of the degree of spectral and/or temporal channel interaction and in other electrical/neural interface factors, it is unclear to what extent variations in these underlying abilities contribute to individual performance (Throckmorton and Collins, 1999). Moreover, there is no consensus as to how to measure these individual
differences, whether through psychophysical or objective means.
It is therefore argued that, in order to differentiate effects of CI signal processing from other factors, it is highly useful to compare results between normal (NH) subjects listening to AMs of CI processing with results obtained from CI users using
equivalent signal processing. This approach has been justified by Throckmorton and Collins (2002) and Laneau et al. (2006), among others. The rationale is as follows: where CI performance and AM performance match, explanations for CI performance can be related directly to model design. More specifically, if an AM which only takes into account CI processing characteristics can predict CI performance, then it follows that CI processing information loss can explain CI user performance. If, however, the model works better if it also incorporates some aspects of the electrical/neural
interface, then it follows that the information loss at the electrical/neural interface must also contribute to the CI performance.
An assumption behind much work in AM research is that a range of different AMs may account for CI user performance so long as those models have the general
properties of CI processing that are perceptually important, e.g. the relatively small number of channels and the absence of temporal fine structure within channels. Almost all AM studies to date have used fixed-channel models and envelope extraction has been via a set of linear IIR filters followed by rectification and
smoothing. However, only some CI systems implement this type of processing while others, notably the Nucleus 24 which is the focus of the present study, use an FFT filterbank and, also, most users use peak-picking strategies such as ACE rather than fixed channel strategies. There is an identifiable need to consider the extent to which the results obtained can be attributed to the specific set of CI processing parameters,
(e.g. pre-emphasis, strategy type, FFT parameters, channel number and channel stimulation rate). Additionally, consideration must be made, specific to AMs
themselves, as to the effects of specific choices of waveform output parameters, (e.g. carrier stimulus) and stimulus parameters, (vowel environment and noise type). It should be noted that each of these specific choices is evaluated in experimental work in the study by comparison with alternatives, with the exception of noise type and input stage processing.
The assessment of consonant feature information transmission provides an opportunity to determine if an AM is predictive of CI user performance. This is because transmission of different consonant features relies on different underlying psychoacoustic abilities and therefore relates to different aspects of signal acoustics. Therefore, it is useful to compare AM performance against CI user performance in a number of ways. First, the pattern of information transmission across consonant features; second, the pattern of effects of background noise across consonant features; third, the pattern of effects of CI processing parameters across features; fourth, the pattern of effects of electrical/neural interface factors across features. If an AM can predict the magnitude and/or pattern of consonant feature transmission as a function of any or all of these variables, then it can be said to have explanatory power in predicting CI performance.
specific research questions can be framed in the context of this more general knowledge gap, e.g. the lack of knowledge about the relative importance of
processing and electrical/neural interface information loss to consonant recognition.
For CI consonant recognition, a consistent finding has been worse place of articulation perception than manner or voicing perception. However, there were identified knowledge gaps in the following areas:
1. Is the pattern of consonant feature transmission in CI users the same in quiet and noise?
2. Is the pattern of consonant feature transmission in CI users the same in vowel environments other than /aCa/?
3. What is the pattern of consonant feature transmission in users of the Nucleus 24 device?
The remaining questions relate to the ability of an AM to predict consonant recognition abilities in CI users:
4. Can an AM accurately predict the pattern of relative consonant feature transmission (in quiet or noise)?
5. Can CI consonant recognition be predicted better by an AM with or without the characteristic shift in perceived pitch associated with CI insertion (referred to as “pitch mismatch”)?
6. Can CI consonant recognition be predicted better by a model incorporating channel interaction and, if so, how much channel interaction is required to optimally match CI user performance?
7. Can variations in channel interaction model variations in CI user performance? 8. Which version of an AM can best predict changes to CI user performance with
changes in channel number?
9. Which version of an AM can best predict changes to CI user performance with changes in stimulation rate?
10.Does choice of AM carrier stimulus have a bearing on the prediction of CI user performance?
Although Laneau et al. (2006) compared performance with a specific device with performance using an AM which incorporated electrical/neural interface features (channel interaction/e.g. spectral channel interaction), the authors assessed different aspects of speech perception than those addressed here. The authors found a
correlation in performance between AM and CI user findings with spectral channel interaction equivalent to 1 mm in the model. It can therefore be hypothesised that consonant recognition in Nucleus 24 users will be best approximated with a model in which channel interaction is equivalent to 1 mm spectral spread.
Experimental hypotheses can either be couched as overall hypotheses or as feature- specific hypotheses. Section 3.3.6 gives a justification for choosing six specific consonant features voicing, place, manner, nasality, fricative and envelope. Ideally, each of the consonant features would have a corresponding hypothesis for each variable in each experiment. A number of feature-specific hypotheses have been put forward in 2.6. More specific hypotheses, including those relevant to processing parameter variables and to specific features, are stated within the context of each experiment in chapters 3 and 4.