An approach to the classification of American English diphthongs

Six diphthongs of American English (/au, aɪ, eɪ, oᴜ, ɔɪ, ju/) were produced by four midwestern American speakers (two male, two female) at two tempos (slow, fast) with differing stress (stressed, unstressed) in two contexts ([b_d], [h_d]). Using a plot of the fundamental frequency and the first three formants derived from linear-prediction-coding (LPC) analysis, the onset and offset of each production was determined. The pattern of formants and fundamental frequency at the onset and offset of diphthongs was used to establish a set of parameters that can classify intended productions of the American English diphthongs in varying stress and tempo conditions with an average accuracy of 93%. Results are also presented for diphthong, target-syllable, and sentence durations. The classification results are discussed with respect to hypotheses concerning the perception of diphthongs.