Browse Prior Art Database

Voice Analyzer

IP.com Disclosure Number: IPCOM000094364D
Original Publication Date: 1966-Oct-01
Included in the Prior Art Database: 2005-Mar-06
Document File: 3 page(s) / 28K

Publishing Venue

IBM

Related People

Buron, RB: AUTHOR [+3]

Abstract

This voice analyzer uses short time spectrum, coherent memory filter, and cepstrum techniques.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 53% of the total text.

Page 1 of 3

Voice Analyzer

This voice analyzer uses short time spectrum, coherent memory filter, and cepstrum techniques.

The speech input signal introduced in 1 is treated in a first coherent memory filter comprising And 2, delay line D1, envelope loop control gate 4, delay line D2, Or 6, and phase shifter phi. Such signal is also treated in a second coherent memory filter comprising And 2, delay line 3, pitch loop control gate 9, logarithm operation circuit 8, Or 6 and phase shifter phi.

An envelope observation time gate 10 and a pitch observation time gate 11 produce the envelope and the pitch of the input speech signal.

In a first time the gating of the analyzer results in the voice being treated in the first coherent memory filter. In this, signal processing is effected by applying the signal to a memory portion of the filter to be stored in it. In this way, the filter is adapted to simultaneously present a multiplicity of stored signals, reproductions of the input signal. Such stored signals are successively delayed by preselected time intervals. These stored signals are then combined to cause wave interference, producing an output signal. Simultaneously with such combination, the phase of each of the multiplicity of stored signals is adjusted in synchronism in search for amplitude peaks in the envelope of the output signal. These amplitude peaks are caused by phase alignment. Each amplitude peak in the output signal is representative of a sinusoidal frequency component of the input signal. Thus the first coherent memory filter produces a spectrum display with a fairly good definition of the formants and shows some ripples representing the pitch information.

The spectrum of speech equals the product of the spectra of the vocal source and of the vocal tract. For pitch and voiced unvoiced detection, information about the vocal source is desired. Therefore the effects of the vocal tract must be separated from the source. This is accomplished by using the cepstrum technique. An appropriat...