A speech signal can be decomposed into the Fundamental frequency and harmon
ics, and the autocorrelation function (ACF) is an effective tool for identi
fying the fundamental Frequency and the harmonics, This paper, thus, explai
ns how ACF harmonic analysis can be applied to speech detection and reconst
ruction when speech communication technologies are used in noisy environmen
ts. The dominant sinusoidal components used for the ACF analysis can be pic
ked out from the short-time Fourier spectrum records of a noisy speech sign
al by using a peak-picking method. Because the number of components usable
for speech reconstruction depends on the signal-to-noise (S/N) ratio, we au
thors developed new methods for peak-picking method and for harmonic sievin
g. The number of components picked our is adjusted frame by frame depending
on the short-time SN ratio, and harmonics are extracted From the short-tim
e Fourier spectrum record by changing the frame length adaptively according
to the fundamental frequency. Consequently, intelligible speech without "m
usical noise" could be reconstructed from noisy speech signals. (C) 2001 Ac
ademic Press.