Automatic Extraction of Formant Frequencies from Continuous Speech
James L. Flanagan · The Journal of the Acoustical Society of America · 1955
An electronic apparatus for extracting formant frequencies from continuous speech is described. The short-time speech frequency spectrum or its second derivative with respect to frequency is sampled rapidly and periodically by scanning the outputs of a thirty-six channel filter set. The time function thus produced is analyzed to determine the spectral maxima or formants. The points in time at which maxima occur are separated and converted into voltages by a further sampling and smoothing operation. These voltage outputs correspond to the formant frequencies as functions of time. A vowel segmenting circuit is incorporated in the input to restrict analysis to the vocalic portions of the input speech if desired. Some preliminary results are presented.