to give users of "O+" calls the of call they wish to make.'cular training set to assess the both template-based and ta collected over the DDD upwards of 75,000 caller -F1 recording equipment.using algorithms currently n accuracy of 99.7% on the these and other experiments will be given in this paper.ion elopment of a speaker-independent speech recognition performs well over dialed-up telephone lines has been a goal of AT&T Bell Laboratories for over a decade [1,21. However, until recently most of our evaluations of recognition systems have been based on laboratory recorded data. These conditions typically consisted of cooperative subjects using local dialed-up lines over a Private Branch Exchange (PBX). Peak signal-to-noise ratios under these conditions generally ranged from 40 -60 dB. Using such local switched lines, the performance of the speech recognition algorithms tested was found to be quite good for a wide range of vocabulary sizes and complexities and for a wide range of talkers.Given the success of laboratory studies, an effort was made to test the viability of our speaker independent, isolated word recognition very large telephone customer populations [3,41. These rmed in 1982 and 1984) were conducted under 'real world' i.e. asking telephone customers to speak their telephone as a series of isolated digits) in a home environment over dialed-up telephone lines while trying to place a normal call. Under these conditions, signal-to-noise ratios of dB and 60 dB were observed. Extensive testing of this r network trials, the recognition algorithms evaluated te-based dynamic time-warping techniques using an lihood distortion measure. Since these tests were introduction of hidden Markov model speech rithms has greatly increased speech recognition procedures [5,61 and new distortion measures [8,9l have also increased database led to a 90% word recognition accuracy.
The art and science ofspeech recognition research have advanced to the point where it is now possible to communicate reliably with a machine via telephone to carry outsimple tasks. However, the creation ofrobust algorithms is only partofthe overall process ofmaking speech recognition technology a commercial reality. In this paper, we present a brief overview ofspeech recognition technology and discuss the implementation ofthe algorithms with digital signal processing chips. In addition, we will show how theory and practice come together in real-world conditions bydescribing a telephone network trial ofautomatic credit card verification using speech recognition technology.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.