In this paper, the AM-FM modulation model is applied to speech analysis, sy
nthesis and coding. The AM-FM model represents the speech signal as the sum
of formant resonance signals each of which contains amplitude and frequenc
y modulation. Multiband filtering and demodulation using the energy separat
ion algorithm are the basic tools used for speech analysis, First, multiban
d demodulation analysis OC IDA) is applied to the problem of fundamental fr
equency estimation using the average instantaneous frequency as estimates o
f pitch harmonics. The MDA pitch tracking algorithm is shown to produce smo
oth and accurate fundamental frequency contours. Next, the AM-FM modulation
vocoder is introduced, which represents speech as the sum of resonance sig
nals. A time-varying filterbank is used to extract the formant bands and th
en the energy separation algorithm is used to demodulate the resonance sign
als into the amplitude envelope and instantaneous frequency signals. Effici
ent modeling and coding (at 4.8-9.6 kbits/sec) algorithms are proposed for
the amplitude envelope and instantaneous frequency of speech resonances. Fi
nally, the perceptual importance of modulations in speech resonances is inv
estigated and it is shown that amplitude modulation patterns are both speak
er and phone dependent. (C) 1999 Elsevier Science B.V. All rights reserved.