Identification of Vowel Formants in Short Segments of Speech
Abstract
A study has been made of vowel formant behavior and a method developed for reliably identifying the first three formants from spectra of isolated 24-msec segments of speech. The problem in formant identification is to distinguish each formant from other spectrum peaks, including other formants. This problem is complicated by extensive overlap of the frequency ranges in which formants characteristically occur, and by irregular variations in formant amplitude, particularly during rapid changes in formant frequency. During the study, formant behavior in several thousand spectra was examined. These data were obtained by fine-resolution spectrum analysis of 100 words spoken by five adult male talkers. The study revealed simple relationships between a formant and the amplitudes and frequencies of the largest and second-largest peaks in its frequency range. These relationships can be reduced to a set of rules that are programmable for execution by computer. Such a program has identified formants with an error rate of less than 5% over a set of 1500 words. [This research was supported by Rome Air Development Center.]
Related Papers
- → Language dependent vowel representation in speech production(2013)30 cited
- → A Comparative Study on the Effects of Age on the Vowel Formants of the Korean Corpus of Spontaneous Speech(2015)6 cited
- → On the role of the fundamental frequency in vowel perception(1988)10 cited
- → An Experiment in Automatic Isolation and Identification of Vowels in Continuous Speech(1974)1 cited
- → Reanalysis of Vowel Formants in the Buckeye Corpus(2020)