Speech Technology at Work IFS

Speech Technology at Work IFS

SPEECH TECHNOLOGY AT WORK Jack Hollingum Graham Cassford Springer-Verlag Berlin Heidelberg GmbH

Jack Hollingum IFS (Publications) Ud 35--39 High Street Kempston Bedford MK42 7BT UK Graham Cassford Logica Energy & Industry Systems U( 64 N ewman Street London WIA4SE UK British Library Cataloguing in Publication Data Hollingum, Jack Speech technology at work. 1. Manufacturing industries. Applications of speech recognition by computer systems 1. Title II. Cassford, Graham, 1957- ISBN 978-3-662-l3014-8 ISBN 978-3-662-l3012-4 (ebook) DOI 10.1007/978-3-662-l3012-4 1988 Springer-Verlag Berlin Heidelberg Urspriinglich erschienen bei Springer-Verlag Berlin Heidelberg New York Tokyo 1988 This work is protected by copyright. The rights covered by this are reserved, in particular those of translating, reprinting, radio broadcasting, reproduction by photo-mechanical or similar means as well as the storage and evaluation in data processing installations even if only extracts are used. Should individual copies for commercial purposes be made with written consent of the publishers then a remittance shall be given to the publishers in accordance with 54, Para 2, of the copyright law. The pubiishers will provide information on the amount of this remittance. Phototypeset by Fleetlines Typesetters, Southend-on-Sea

Contents Chapter 1 Introduction 1 Speech recognition 4 Speech output 7 What makes speech technology worth investigating? 8 Limitations 10 Chapter 2 Recognising speech 13 What is speech? 15 Speech recognisers today 20 Vocabulary size 22 Chapter 3 Reproducing speech 25 Speech compression 26 Speech synthesis 27 Dedicated speech chips 31 Text to speech 32 Chapter 4 The Scope for speech technology 37 Voice messaging 38 Telephone voice response: keypad systems 40 Telephone voice responses: voice input 47 Industrial applications 49 Surgical applications 55 Educational aids 56 Language translation 58 Aids to the handicapped 59 Consumer products 60 Speech to text 61

Chapter 5 Human factors 65 The importance of dialogue design 69 Choosing the vocabulary 72 Training procedures 73 Quality of voice response 74 Chapter 6 How to get started 77 Taking advice versus working independently 78 Where to look for applications 80 Economic advantages 81 Comparing speech with other methods 83 Performance evaluation of recognition systems 83 Chapter 7 Industrial case studies Jaguar Car 86 Caterpillar Tractor 96 Austin Rover 101 Rolls-Royce 113 Ford Distribution Centre 118 85 Chapter 8 The future 125 Telephone voice input and response 127 Language translation 129 Improvements in recognisers 130 Recogniser assessment 132 Text to speech 134 Synthesis by analysis 135 Speech coding 136 Appendix Systems available 139

Graham Cassford gained a BSc Hons from Portsmouth Polytechnic in 1980 and a PhD from Reading University in 1983. He began his career with British Robotic Systems Ltd as a software engineer concerned with the development of Industrial Vision Processing Systems. In 1985 he joined Logica Energy and Industry Systems Ltd as a senior consultant within the Industrial Systems Group of the company. He is responsible for the design and implementation of shop-floor computer systems, specialising in the application of speech- and vision-based technologies. Jack Hollingum is a distinguised technical author and journalist, and a former editor of Sensor Review magazine. His previous publications include Machine Vision: The Eyes of Automation (1984); The Machine Vision Sourcebook (1986); Implementing an Information Strategy in Manufacture (1987); and Implementing Total Quality (1987), all published by IFS.

Preface Speech technology - the use of speech as a means of sending information to, and receiving information from computer systemshas been in use as a research tool for many years. Only recently has it begun to move out of the laboratory and into commercially worthwhile applications, first with compressed and synthesised spoken messages, then with computer recognition of spoken messages, and today with diverse applications involving both recognition and reproduction of human speech. We have written this book because we believe the technology has now advanced to the point where many more applications of voice recognition and response are both feasible and economically attractive. Computers that can understand everyday speech are still a distant prospect, but provided the limitations of present day equipment are clearly understood there is much that can be achieved with it. Our aim is to show, in non-technical language, what is now possible with the help of speech technology. The text includes many examples of current applications in industry, commerce and other fields, and we have selected five current industrial applications combining speech recognition and response for more detailed attention. Industrial cases have been chosen both because we see industry as an important growth area for speech applications in the next few years, and because it presents some of the greatest difficulties in speech recognition - if you can make it work in industry, then you can make it work almost anywhere. Included in the book as an appendix are brief details of companies offering equipment or services connected with speech technolgy. It

is based on information supplied by those compaies and makes no claim either to completeness or to endorsing the products and services mentioned. Nevertheless we hope that it will be of some help to anybody wishing to explore further the opportunities for introducing speech technology. A book like this inevitably draws on the knowledge and experience of a great many people, and it would be impossible to name or even to identify all of them. Several people, though, have given personal advice and help which we would like to acknowledge without in any way implicating them in any errors which may have crept into the text. We have drawn on the experience of a number of people in Logica. In particular Dr Jeremy Peckham, who is an international authority on speech recognition, has given a great deal of advice on the subject-matter of the book, as well as reading and commenting on the final draft. Chapters 2, 5, 6 and 8 in particular make considerable use of his expertise. Chris Wheddon, head of speech and language processing at the British Telecom Research Laboratories, gave a fascinating insight into work which should soon bear fruit in a number of important areas. Raj Gunawardana of the radio and speech systems group at Texas Instruments was helpful in reviewing developments in voice response and speaker independence. For information relating to the industrial case studies we are deeply indebted to Jaguar Cars Ltd., Caterpillar (UK) Ltd.; to Kapul Gill of Austin Rover for his contribution on the Rover Inspection system; to Alan Patrick of Rolls-Royce for detailed information on the voice-controlled inspection system at the company's Precision Casting Facility; and to Ford Motor Company and Computer Gesellschaft Konstant for information on the Cologne Distribution Centre. Other application details have come from many sources, but particularly from the companies which have also supplied details of their products and services - our grateful thanks to them.