Browse Prior Art Database

Fast Digital Filterbanks for Speech Processing

IP.com Disclosure Number: IPCOM000079524D
Original Publication Date: 1973-Jul-01
Included in the Prior Art Database: 2005-Feb-26
Document File: 2 page(s) / 38K

Publishing Venue

IBM

Related People

Bakis, R: AUTHOR [+2]

Abstract

This is a cascaded arrangement of speech processing digital filterbanks in which the sampling frequencies for digital filters in the lower octaves are reduced, thereby speeding up the signal processing computations and avoiding the precision problems which otherwise would arise in the lower octaves, if a uniform high-sampling rate were used for all octaves.

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 69% of the total text.

Page 1 of 2

Fast Digital Filterbanks for Speech Processing

This is a cascaded arrangement of speech processing digital filterbanks in which the sampling frequencies for digital filters in the lower octaves are reduced, thereby speeding up the signal processing computations and avoiding the precision problems which otherwise would arise in the lower octaves, if a uniform high-sampling rate were used for all octaves.

Implementations A and B are shown. In Implementation A the bandpass filters are grouped by octaves. The input signal is fed to the highest-octave bandpass filters and also to a first low-pass filter, which excludes the highest- octave components and passes the lower-octave components to the second- octave group of bandpass filters and also to a second low-pass filter. Each low- pass filter, in turn, serves to transmit only the frequencies that are below the octave of the group with which it is associated.

In Implementation B, a low-pass filter is used in place of each of the bandpass filters of Implementation A. The high-pass output is created in each instance by subtracting the low-pass output from the input to the respective low- pass filter. Implementation B requires fewer operations than Implementation A, but is somewhat less accurate.

In both forms A and B there is a progressive reduction in the required sampling rate for the filters, as filtering proceeds towards the lowest-frequency end of the system. Thus, the sampling rate is kept commensurate with the fr...