Browse Prior Art Database

Extended Linear Prediction Coding for Speech Analysis

IP.com Disclosure Number: IPCOM000035355D
Original Publication Date: 1989-Jul-01
Included in the Prior Art Database: 2005-Jan-28
Document File: 2 page(s) / 39K

Publishing Venue

IBM

Related People

Feig, E: AUTHOR

Abstract

A linear prediction coding (LPC) technique which provides additional parameters beyond predictive coefficients and variance is disclosed. The additional parameters are the expected values of noise, and it is suggested that noise correlates with the future signal as a voice signal depends on driving noise.

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 55% of the total text.

Page 1 of 2

Extended Linear Prediction Coding for Speech Analysis

A linear prediction coding (LPC) technique which provides additional parameters beyond predictive coefficients and variance is disclosed. The additional parameters are the expected values of noise, and it is suggested that noise correlates with the future signal as a voice signal depends on driving noise. The preferred technique comprises: a) obtaining a signal, digitizing the signal and autocorrelating samples of the signal to provide values of an autocorrelation function of order P for at least 2P + L + 1 equally spaced lags; b) sending the first 2P + 1 lag values to an LPC parameter generator which yields parameters consisting of predictive coefficients and the variance; and c) computing the expected noise value parameters by multiplying the predictive coefficients and the last L + P + 1 lag values in a matrix- vector multiplier. The provided technique captures some of the information which is typically lost when passing from the Fourier Transform to the power spectrum in analyzing speech.

Parameters indicative of an individual's speech which will identify the individual may be obtained by performing a standard LPC parameter extraction on a short-time sample of that speech. The sampled values xj of the speech satisfy a difference relation: (1) where ao = 1, the LPC coefficients a1,...,ap are to be determined, and wn are sampled values of a white noise process such that the expected values E wnxn+1 = 0 for 1...