Explicit modeling of vowel coarticulation in continuous speech recognition

An ongoing study is reported of all sixteen of the American English vowels using subsets of the DARPA acoustic-phonetic database. Formants are obtained and normalized for each talker´s formant range based on one sentence. The resulting formant tracks are smoothed using splines and sampled at nine equally spaced points in time within vowel-centered triphone regions. Triphones with semivowels in them are clustered separately. These formant values are k-means clustered using subsets of the sampled formant values. The additional supervised training is done using other parameters, including duration. The resulting clusters are used as a classifier on the basis of the modified Euclidean distance from the cluster centers. This results in approximately 80% first choice vowel recognition of the outer edges of the vowel quadrilateral. Stressed vowels were found to have spectra which statistically were no more stable than unstressed vowels