A speech analysis system in which segments of digitized speech are transformed into amplitude spectrums. For the voiced/unvoiced decision use is made of the peak value or spectral intensity in each amplitude spectrum. Basically a voiced decision is made when the spectral intensity increases monotonically over several segments by more than a given factor. An unvoiced decision is made if the spectral intensity drops below a given fraction of the maximum spectral intensity in the current voiced period. Refinements in the decisions are made by the use of fixed and adaptive thresholds.