Browse Prior Art Database

Improved Vowel Separation For Speech Recognition Applications

IP.com Disclosure Number: IPCOM000097332D
Original Publication Date: 1962-Oct-01
Included in the Prior Art Database: 2005-Mar-07
Document File: 2 page(s) / 32K

Publishing Venue

IBM

Related People

Dersch, WC: AUTHOR

Abstract

This device is for improving speech recognition techniques and the transmission of speech information more efficiently. The method distorts the input speech wave by running it through a number of pass filters in tandem with phase shifters. It then detects these different wave components. Differences between syllables having similar wave forms are accentuated.

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 81% of the total text.

Page 1 of 2

Improved Vowel Separation For Speech Recognition Applications

This device is for improving speech recognition techniques and the transmission of speech information more efficiently. The method distorts the input speech wave by running it through a number of pass filters in tandem with phase shifters. It then detects these different wave components. Differences between syllables having similar wave forms are accentuated.

The system involves the selection by Pass-Filters of multiple frequencies F(r), 2F(r), 4F(r), etc., which are frequencies characteristic of speech components that are difficult to discriminate conventionally. The result is that the sounds to be discriminated are made to exhibit a more unique energy output and thus be more readily distinguished.

The input speech wave is presented to microphone M and then amplified at Amplifier A. It is subsequently presented to a parallel series of filter lines F(1), F(2) and F(n). These filters are selected so as to pass selected critical frequencies of the input wave. The output of each band pass filter is fed into an individual phase shifter PS capable of shifting a reference frequency from 0 - 360 degrees. The outputs of these phase shifters are then fed to amplifiers A(1)... A(n) individually. The audio wave is then summed and presented to envelope demodulator means ED+ and ED-. The result is an output envelope at 0 having asymmetry envelope characteristics which can then readily be analyzed.

Envelope detector...