Mining approach on the acoustic parameter analyses of sustained Turkish vowels

This study analyzes and evaluates the acoustic parameters of sustained Turkish vowels. Recordings were taken from 17 volunteer adults, who expressed that they did not demonstrate any pathologies. Volunteers uttered /a/, /e/, /i/, /o/ and /u/ vowels twice, the second recordings being made approximately two weeks after the first. Vowels were recorded and analyzed using Kay Elemetrics Multi-Speech (MS) analysis software. The acoustic parameters analyzed were jitter, shimmer, pitch period and HNR (harmonic-to-noise ratio). The results obtained from the MS software were evaluated against normative values obtained from healthy people. After the parameter analyses, all acoustic parameters under investigation for each vowel, obtained from the same volunteer, (i.e. jitter, shimmer, pitch period, HNR) were examined by a visual data mining method. The results indicate that there is no significant difference between the first and second recordings for the parameters under investigation. In addition to this important finding, it is revealed that just one of the vowels, except /u/, is enough for any decision about someone's recordings.