Applying pattern recognition to formant trajectories: A useful tool for understanding African American English (AAE) dialect variation

Few studies have focused on the acoustic-phonetic characteristics of African American English (AAE) which distinguish this dialect from Standard American English (SAE), particularly for vowels and sonorant consonants. This study investigated whether formant dynamics from short, sonorant portions of speech are sufficient to distinguish AAE and SAE dialects. Seven female speakers, four SAE and three AAE, from the Lansing, Michigan area, were selected from a corpus of 30-45 minute sociolinguistic interviews. Target portions of speech consisting of a V or VC sequence (C = /n/, /m/, /l/, /r/) were identified from contexts selected to control for coarticulation. First (F1) and second (F2) formant values were extracted from randomly selected tokens at points 19%, 56%, and 81% of the duration through the demarcated speech portions. Pattern recognition techniques were examined to differentiate tokens of the two dialects based on formant trajectories as feature vectors. The results revealed that formant dynamics of the selected contexts are acoustically informative enough to differentiate groups of SAE from AAE speakers. A near-perfect classification of some contexts was also achieved by applying support vector machines to the formant trajectories. These findings highlight the usefulness of incorporating pattern recognition techniques for understanding acoustic variation due to dialect.