Formant tracking using hidden Markov models

This paper describes an approach to formant tracking based on hidden Markov models and vector quantization of LPC spectra. The overall formant tracking problem is decomposed into two sequential subproblems- detection and estimation. Formant detection involves making a binary decision about the presence of a formant for each input frame. Formant estimation is concerned with obtaining a numerical formant frequency for each frame in which a formant is detected. Both steps involve finding an optimal state sequence for a hidden Markov model using the Viterbi algorithm. The method has been applied to the problem of F2tracking and a preliminary evaluation performed using the Texas Instruments connected digits database. The F2detector exhibited false alarm and missed event rates of 8% and 5%. The average absolute and root-mean-square F2estimation errors were 56 Hz and 83 Hz.