This paper treats the problem of distinguishing speech from voiceband
data. It presents a discrimination function based upon the sign of the
autocorrelation at lag two of the incoming signal and the second-orde
r moment of the complex low-pass signal. The algorithm has been applie
d to many types of voiceband data signals and, within a window of 32 m
s, correctly classifies them. For the limited number of speech signals
available, the amount of misclassification of speech as voiceband dat
a was observed to be about 1%.