한국어 Matrix 문장 검사 도구 평가

Evaluation of the Korean Matrix Sentence Test: Verification of the List Equivalence and the Effect of Word Position

Purpose: Multilingual Matrix Sentence Tests have been developed for reliable, precise, and internationally comparable sentence-innoise tests. The lists of the Korean Matrix Sentence Test have been constructed and optimized in the previous study. The sentences are generated by a base matrix of 50 Korean words (10 names, 10 adjectives, 10 objects, 10 numerals, and 10 verbs). As suggested by International Collegium of Rehabilitative Audiology working group, this study conducted evaluation tests to verify the list equivalence and the effect of word position at a fixed signal-to-noise ratio (SNR). Methods: For experiment I, fifteen normal-hearing (NH) listeners participated. As a material, the Korean Matrix sentences optimized and level-adjusted were used. Twelve test lists of ten sentences were combined to six test lists of twenty sentences (double lists) such that each listener was tested with six double-lists in an open-set format. The percentage of correct responses was measured based on word scoring at fixed SNRs of -10, -8, -6, -4 dB SNRs, with a constant noise level of 65 dB sound pressure level. List-specific intelligibility functions for each list were derived by fitting the percent-correct scores to the logistic regression function, yielding speech reception threshold (SRT) at which 50% intelligibility and also the slope at the SRT across listeners. For experiment II, thirty hearing-impaired (HI) subjects participated and all of them have used hearing aids bilaterally. The aided puretone threshold average of HI listeners was 35.9 dB HL. The materials of the Korean Matrix sentences were presented through loudspeakers while wearing listener’s hearing aids. Based on word scoring, the listeners’ recognition score was measured in four listening conditions (quiet, 6, 3, 0 dB SNRs). Results: In experiment 1, the mean SRT and mean slope across the six test lists were -9.56 dB SNR and 14.03 %/dB, respectively. Across the fixed SNRs, mean difference of SRT was within 1 dB across lists. Based on word scoring, the NH listeners were found to poorly recognize the word at the first position than words at other positions. The mean SRT and slope of the Korean Matrix test were comparable to results of other multilingual Matrix tests. In experiment 2, the results of HI listeners depended on the listening condition (quiet, 6, 3, 0 dB SNR). Regardless of condition, HI listeners poorly recognized words at the first position than words at other positions. The recognition score obtained from the closed-set format was greater than from the open-set test format. Conclusion: The equivalence in intelligibility was confirmed by the results of the Korean Matrix sentence recognition at the fixed SNRs. The results obtained from Korean Matrix test were comparable to the results of other multilingual Matrix tests. The optimized and evaluated Korean Matrix Sentence Test can be used for accurate and reliable speech intelligibility in noise measurements.