A Speech Analyzer and Synthesizer

Publisher Website
Google Scholar
Abstract
This paper describes work on a resonance type of Vocoder that was started just before the war and discontinued because of war work. The project has not been renewed but the first results obtained are reported now since there is considerable interest at the present time in this type of analysis and synthesis and in the circuits used. The circuits were designed to extract the important characteristics of speech by determining the frequency and magnitude of the principal vocal resonances as continuous functions of time. Speech was split into three bands: 300 to 1100, 900 to 3000, and 3000 to 8000 cycles. In each band the frequency of axis crossings in the wave form and the total energy were measured by suitable circuits. In addition, the fundamental pitch of the cord tone and the energy in the band from 40 to 400 cycles were measured. These parameters were transmitted in eight 40 cycle bands to a synthesizing circuit which reconstructed the speech. The synthesizer contained two signal sources, a cord tone with controlled pitch and a wide band noise. These two signals were passed through resonant networks having controlled peak frequencies. The frequency responses of the networks simulated the characteristics of speech formants and the peak frequencies were shifted by variable inductance coils controlled by the axis crossing counting devices in the analyzer. Data are shown on the performance of the analyzer and the combined analyzer and synthesizer.

This publication has 0 references indexed in Scilit: