Chapter 2. Background
3.3 Methodology
3.3.5 Testing regime
For all experiments, subjects were seated in a double-walled soundproof booth. Stimuli in the AM experiments were presented monaurally to the subjects via an ER-3 insert earphone connected to a PC with a Sound Blaster sound card or via the sound field for the CI user experiment. Insert earphone presentation was used for AM experiments in order to minimise the effect of the pinna/outer ear transfer function on the stimuli. Monaural presentation was used as this mimics the normal listening condition for the majority of CI users. Stimuli were presented to subjects via routing
of the signal into an adjacent soundproof booth. A touch screen was used for visual presentation of response options and to code subjects’ responses. The ear to which the sounds were presented was alternated between subjects.
Testing was undertaken using the Praat (version 4.1) speech analysis and testing software, developed by Paul Boersma and David Weenink of the Institute of Phonetic Sciences at the University of Amsterdam (www.fon.hum.uva.nl/praat/). Stimulus presentation was controlled using scripts developed from the Praat (v 4.1) speech analysis toolkit. The software was designed to enable speech analysis but also enabled code to be generated to run and score speech perception experiments. The Praat code, or programme, randomised presentation of stimuli kept in the same source folder as the Praat script. The script also generated a graphical-user interface that was used on a touch screen to record subject responses. Each stimulus presentation required the subject to click on an icon on the screen before the next stimulus was presented. The same set-up was used in all subsequent experimental work reported in this thesis. The only difference between the three AM experiments and the CI user experiment was stimulus presentation and level: for the AM experiments, stimuli were presented by monaural insert earphone, whereas for the CI user experiment, stimuli were presented via sound field presentation as described in 5.
An important aspect of the experimental approach used was the nature of acclimatisation to stimuli. Davis (2004) noted that relatively brief familiarisation with noise-vocoded speech, e.g. of 20 minute or less, led to improved performance with sentence recognition. Some authors have noted that there is an intial “pop-out” effect of vocoded speech, e.g. when a stimulus is defined (e.g. the listener is exposed to the AM stimuli, then told what the word or speech sound is, the salience of the stimulus “pops out”). However, other authors, notably Rosen et al. (1999) have noted that considerable acclimatisation time is needed to achieve optimal performance for NH subjects listening to AMs. One of the questions for this study is whether sufficient acclimatisation would occur over a relatively short time period to yield valid results. The approach taken in the first experiment was to present all 20 stimuli on the touch- screen, each labelled (for example, /idi/ was labelled as “d”) and allow the subject to listen to each stimulus as many times as s/he wished prior to testing proper. For
experiments 2 and 4, there were a large number of AM conditions; consequently, it was impractical to allow this task prior to every listening condition. Therefore, the self-directed acclimatisation process took place for only one AM condition in quiet, and then one AM condition in noise, randomised across the subjects in each experiment. In practice, the self-directed acclimatisation process took between 5 and 10 minutes. A check of identification of four of the tokens was undertaken at this point to determine whether subjects had acclimatised sufficiently to AM stimuli A striking finding was that subjects stated that the stimuli were much clearer after only this short acclimatisation period. The results (see chapters 3 and 4) also indicated that this approach to AM acclimatisation yielded valid results.
For all 4 experiments, quiet listening conditions preceded noise-contaminated
listening conditions, e.g. the design of the experiments was not randomised across the quiet vs. noise contrast. The rationale for adhering to this was the desire to provide further acclimatisation to the model when undertaking noise-contaminated listening conditions via testing in the quiet listening conditions (given the modest amount of acclimatisation time given in the first place). This meant that the effect of noise may have been diluted by consistent exposure to quiet AMs prior to noise-contaminated AMs and that, as a consequence, the possibility of a type II error (with respect to the noise variable) was increased. However, it also meant that the possibility of a type I error for the noise variable was minimised and, where significant effects of noise were obtained, these were more robust.
For experiment 1, stimuli were first presented in the “unaltered” condition, then “quiet AM”, then “AM +10 dB SNR”, then “AM+5 dB SNR” then “AM 0 dB SNR”. For experiment 2, the 8 listening conditions (2 vowel environments * 2 pitch shift conditions * 2 carrier stimulus types-see 4.1.2 for further details) were randomised across subjects and in each case quiet then noise variant of the listening condition was presented. For the CI user experiment, the three MAP conditions were randomised across the 9 subjects, but again within each of these the quiet presentation was undertaken first, followed by the noise-contaminated condition . For experiment 4, testing was first undertaken in the “unaltered” condition. Following this, the three channel interaction conditions were randomised, then the three MAP conditions within each channel interaction, but, again, first the quiet then the noise-contaminated
version of each listening condition were presented. Randomisation was achieved via coding of each listening condition and using a random number generator implemented in Microsoft Excel.