Telephony channel simulator for speech recognition application
First Claim
Patent Images
1. A method for training a speech recognition processor to respond to speech obtained from telephone systems, comprising the steps of:
- inputting a speech data set to a speech recognition training processor, said data set having a bandwidth higher than a telephone bandwidth;
decimating said inputted speech data set in said training processor to obtain a decimated speech data set having said telephone bandwidth;
applying a bandpass digital filter to said decimated speech data set in said training processor, said filter characterizing transmission characteristics of telephone equipment, for obtaining a filtered speech data set;
rescaling the amplitude of said filtered speech data set in said training processor, so that the maximum dynamic range of said filtered speech data set matches the maximum dynamic range of uncompanded telephone speech, to obtain a rescaled speech data set;
modifying said rescaled speech data set in said training processor, with quantization noise representing companding and uncompanding a speech signal in a telephone system, to obtain a modified speech data set;
inputting said modified speech data set into a hidden Markov model speech recognition processor to train statistical pattern matching data units;
performing speech recognition on voice signals from a telephone system with said speech recognition processor.
0 Assignments
0 Petitions
Accused Products
Abstract
A telephony channel simulation process is disclosed for training a speech recognizer to respond to speech obtained from telephone systems. An input speech data set is provided to a speech recognition training processor, whose bandwidth is higher than a telephone bandwidth. The process performs a series of alterations to the input speech data set to obtain a modified speech data set. The modified speech data set enables the speech recognition processor to perform speech recognition on voice signals from a telephone system.
239 Citations
7 Claims
-
1. A method for training a speech recognition processor to respond to speech obtained from telephone systems, comprising the steps of:
-
inputting a speech data set to a speech recognition training processor, said data set having a bandwidth higher than a telephone bandwidth; decimating said inputted speech data set in said training processor to obtain a decimated speech data set having said telephone bandwidth; applying a bandpass digital filter to said decimated speech data set in said training processor, said filter characterizing transmission characteristics of telephone equipment, for obtaining a filtered speech data set; rescaling the amplitude of said filtered speech data set in said training processor, so that the maximum dynamic range of said filtered speech data set matches the maximum dynamic range of uncompanded telephone speech, to obtain a rescaled speech data set; modifying said rescaled speech data set in said training processor, with quantization noise representing companding and uncompanding a speech signal in a telephone system, to obtain a modified speech data set; inputting said modified speech data set into a hidden Markov model speech recognition processor to train statistical pattern matching data units; performing speech recognition on voice signals from a telephone system with said speech recognition processor. - View Dependent Claims (2, 3, 4, 5, 6, 7)
-
Specification