Design of an Experimental-Model Formant Vocoder

Abstract

This paper discusses the design of an experimental-model formant vocoder that employs partial spectral-pattern-matching for the tentative identification of the formants of each short-time spectrum. The speech signal is converted to a sequence of short-time spectra by means of a 55-channel bank of bandpass filters and a sequential sampling switch operating at 100 short-time spectra per second. All of the formantlike energy distributions of each short-time spectrum are identified by matching the spectrum against a set of standard formant patterns. From this set of potential formants, three are selected by rules and are identified as formants one, two, and three. The data describing the three formant frequencies, intensities, and excitation are transmitted to a parallel-formant type of synthesizer. A tape demonstration of the system is presented. [This work was supported by the U. S. Army Electronics Research and Development Laboratory, Fort Monmouth, N. J., and General Dynamics Electronics.]