Evaluation of a novel phrase-based speech-recognition test using synthetic speech

OBJECTIVE: Compilation and evaluation of test lists for the synthesised speech test "Oldenburg Phrases" (OLPHRA), consisting of semantically meaningful phrases with the fixed structure article - adjective - noun - infinitive. DESIGN: Lists of 20 phrases were created based on different properties (e.g., speech-recognition thresholds (SRT), phoneme classes). From 1000 lists, 21 were randomly selected to analyse perceptual equivalence and training effects. Furthermore, OLPHRA was compared to established speech tests. STUDY SAMPLE: The study involved 30 young participants with normal hearing (YNH) and 14 older participants with hearing impairment (OHI). RESULTS: Perceptual equivalence of 21 lists was demonstrated for the YNH group, with an average SRT of -9.0 dB SNR and a maximum difference of 1.2 dB SNR between lists. The mean slope of the psychometric functions was 14 percentage points per dB. The Speech Intelligibility Index indicated acoustical equivalence for the remaining lists. All lists were equivalent in terms of phoneme classes, and small training effects were observed. For the OHI group, the SRT for OLPHRA did not differ significantly from established tests, and all tests showed high test-retest reliability (intraclass correlation coefficients ≥ 0.9). CONCLUSION: OLPHRA fulfils the requirements of ISO 8253-3 and promises versatile use in audiology.