The development of a new technique for the normalisation of the University of Canterbury adaptive speech test- filtered words (UCAST-FW).
Degree GrantorUniversity of Canterbury
Degree NameMaster of Audiology
Low-pass filtered word tests, in which a speech sample is degraded using a low-pass filter (LPF), are one class of low-redundancy test commonly used in the diagnosis of auditory processing disorder (APD). Due to the high level of redundancy within the auditory system and in spoken language, a normal listener is able to fill in the missing speech information and achieve auditory closure even when the speech signal is degraded. The ability to achieve auditory closure is compromised in individuals with APD, which allows filtered speech tests to be used in the diagnostic assessment of APD. One example of this type of test is the University of Canterbury Adaptive Speech Test – Filtered Words (UCAST-FW; O’Beirne, McGaffin and Rickard, 2012). However, the validity and reliability of speech tests are affected by a number of factors, including the homogeneity of the word list. While the UCAST-FW is sensitive enough to discriminate between children with and without APD (Rickard, Heidtke & O’Beirne, 2013), the large variance in the spectral content of its individual test items has resulted in it being somewhat heterogeneous in regards to recognition performance under the same levels of filtering. This creates inherent vulnerabilities within the sensitivity and specificity of a diagnostic test, with increased interitem variability and reduced inter-patient variability. The present study aimed to compensate for differences in word recognition performance among each word in the UCAST-FW by adjusting the level of filtering such that each word is equally difficult. The performance of 61 English speaking adult listeners with normal hearing was examined on their ability to discriminate speech items both before normalisation (n = 30) and after (n = 31). Psychometric functions (percentage correct vs. LPF frequency) were generated for each word, and were used to calculate relative adjustments for the level of low-pass filtering. These adjustments were performed using a novel method of normalisation that adjusts the levels of low-pass filtering relative to the average performance and takes into consideration the psychometric slope of each of the test words rather than just the midpoint of the function. Results from this study show this normalization technique was successful in achieving a more homogenous word list for both open and closed set testing paradigms, relative to the prenormalisation testing, as indicated by a more normally distributed cluster of psychometric functions.