Speaker-identification with real time formant extraction
H. J. Reitboeck, T. Brody, D. Thomas · 2005
The Westinghouse speaker identification system uses frequency-locked loops for the extraction of the first and second formant of voiced phonemes. The analog control voltage of the loop circuits (digitized at 8 bits/sample) represents characteristic formant features at a bit rate that is low enough for subsequent real time processing by an Intel 8080 microprocessor. The system determines for each formant feature a statistical weight factor that depends on how well a feature reproduces in repeated utterances, and finally assigns to each speaker a global accept/reject threshold.