Speech analysis/Synthesis based on perception

A system is described in which a bank of bandpass filters is used to analyze signals. Filter characteristics are chosen to match the empirical critical bands of human auditory perception. Each bandpass filter is followed by a detector. A Generalized Short-Time Fourier Transform (GSTFT) magnitude analysis is used to implement the Filter/Detector bank. An efficient recursive implementation is described. Synthesis equations are presented which allow signal reconstruction from the GSTFT magnitude. An example is given in which a sinusoidal signal is analyzed and synthesized by this method.