An integrated dialect analysis tool using phonetics and acoustics

This study aimed to verify a computational phonetic and acoustic analysis tool created in the MATLAB environment. A dataset was obtained containing 3 broad American dialects (Northern, Western and New England) from the TIMIT database using words that also appeared in the Swadesh list. Each dialect consisted of 20 speakers uttering 10 sentences. Verification using phonetic comparisons between dialects was made by calculating the Levenshtein distance in Gabmap and the proposed software tool. Agreement between the linguistic distances using each analysis method was found. Each tool showed increasing linguistic distance as a function of increasing geographic distance, in a similar shape to Seguy's curve. The proposed tool was then further developed to include acoustic characterisation capability of inter dialect dynamics. Significant variation between dialects was found for the pitch, trajectory length and spectral rate of change for 7 of the phonetic vowels investigated. Analysis of the vowel area using the 4 corner vowels indicated that for male speakers, geographically closer dialects have smaller variations in vowel space area than those further apart. The female utterances did not show a similar pattern of linguistic distance likely due to the lack of one corner vowel /u/, making the vowel space a triangle.