Second International Conference on Spoken Language Processing (ICSLP'92)
Banff, Alberta, Canada
A modification of the Distinctive Features Difference (DFD) test  was developed as part of a speech perception test battery, primarily for use in testing new hearing aid circuitry and noise reduction algorithms. The DFD(m) test requires listeners to identify each of 21 consonant sounds, spoken by two male and two female talkers, presented in an /ACd/ context against a noise background. Identification errors can be scored in terms of either the overall percent-correct item identification or the number of feature differences between each target sound and response. Performance-intensity functions were obtained with a 70 dB(A) background noise masker that was spectrally shaped to match the long-term average spectrum of the 84 test items. Progressive low-pass and high-pass filtering of the speech was used to obtain the crossover frequency of the Articulation Index importance weights for the combined 4-talker speech materials. The crossover frequency was 2170 Hz, higher than that previously found for other sets of nonsense syllables. Articulation Indices were computed using three methods and their relatively suitability for application to the modified DFD test was investigated.
Bibliographic reference. Cheesman, Margaret F. / Lawrence, Shelly / Appleyard, Allison (1992): "Performance on a nonsense syllable test using the articulation index", In ICSLP-1992, 1123-1126.