Role of Different Spectral Attributes in Vowel Categorization: the Case of Udmurt

The present study examines the difference between categorization and goodness ratings in Udmurt (Finno-Ugric language) using different sets of spectral attributes. The tendency to have two areas of /u/ vowels was observed in languages with unrounded non-front closed vowels during the TURVOTES project. Our study explores whether this can be due to the different acoustic attributes used in the identification and goodness ratings of these vowels. The identification and goodness-rating of Udmurt close vowels confirmed the observation. The model using only formants was not significant for identification data, but did explain the goodness rating data. The spectral moments explained both identification and goodness ratings.