Evaluation of hearing loss simulation using a speech intelligibility index


ARIÖZ U., GÜNEL KILIÇ B.

TURKISH JOURNAL OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCES, cilt.24, ss.4193-4207, 2016 (SCI İndekslerine Giren Dergi) identifier identifier

  • Cilt numarası: 24 Konu: 5
  • Basım Tarihi: 2016
  • Doi Numarası: 10.3906/elk-1411-135
  • Dergi Adı: TURKISH JOURNAL OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCES
  • Sayfa Sayıları: ss.4193-4207

Özet

Hearing loss simulation (HLS) is a useful tool for hearing studies, since finding large numbers of hearing impaired volunteers with various levels of hearing loss is usually a problem. Correct simulation of the hearing impaired ear should take into account different suprathreshold effects, such as reduced frequency selectivity, reduced audibility, and loudness recruitment. Although these effects can be implemented using various algorithms, so far they have not been evaluated in detail taking into account subject dependency, effect of noisy environment, and phonetic characteristics of test words. In this study, for the purpose of evaluating the HLS, a speech intelligibility index (SII) was used as the objective measure and a modified rhyme test (MRT) was used as the subjective measure. Hearing loss of 12 subjects was simulated using various stimuli and 36 subjects with normal hearing were used as a control group. The factors that affect the performance of HLS algorithms were initially determined using the SII metric. Three other factors (sex of speaker, background noise level, and characteristics of the words) were tested both by the SII metric and the MRT. The lists contained word groups designed according to Turkish phonetic characteristics. In addition, the results of the MRT were compared with the results of the SII for both original and simulated sounds. According to the mean values and correlation analysis, both measures gave similar and reliable results for the HLS (58.40 and 57.37 for MRT mean values of unprocessed and simulated sounds, respectively; 0.23 for MRT mean values of unprocessed and processed sounds; 58% and 74% correlation coefficients for MRT and SII, respectively). When statistical significances of the measures were taken into account, the MRT gave more reliable results than the SII. While sounds were simulated in a satisfactory manner in both the noise-free environment and noisy environment for the MRT, similar results for the SII were obtained only in the noise-free environment. Because of high sensitivity of the SII to noise, the results were not satisfactory for the noisy cases. After the simulation, sex of the speaker and test list factors were found to be significant with the noise factor for the MRT.