image

PREFACE TO THE 2011 EDITION

0.1 WHY WE CREATED A NEW EDITION

Technology moves at a dizzying pace; however, progress can actually seem quite slow in any area that we are deeply involved in. Conference proceedings are filled with incremental advances over previous methods, and entirely novel (and successful) approaches to speech and audio processing are rare. But a lot can happen in a decade, and it has. In addition to quite new methods, there are also many ideas that had not really been refined enough to show progress in the 1990s, but which now are in common use. For instance, Maximum Mutual Information methods, which were developed for ASR many years ago and were briefly described in the previous edition of this book, was significantly refined in the last decade, and the newer versions of this approach are now widely used. Consequently, we devoted new sections of this revision to MMI (and related methods like MPE).

These advances might have been sufficient to warrant an update of our textbook, but there were other reasons as well. A decade of teaching with the book has revealed a number of bugs and deficiencies, and a new edition affords us the opportunity to correct them. For instance, the previous version had nothing about sound source separation, an area that has received considerable attention in the last decade. Approaches to the coding, transcription, and retrieval of music are also now significant areas of audio signal processing, and were not originally covered in the book.

Last, and not least, the new edition has the benefit of a fresh look at the overall subject from our new co-author, Professor Dan Ellis from Columbia University. This hand-off is a key step in keeping the text current.

As with the previous edition, we've attempted to keep the overall style consistent, focusing on what we think is essential, and leaving many details for other publications. we hope that this choice has helped to make the text useful for many readers.

0.2 WHAT IS NEW

As noted above, we have edited and modified many of the chapters, but we also have added entirely new ones. These are:

  • Acoustic model training: further topics – MAP and MLLR adaptation methods, and on MMI and MPE discriminant training (Chapter 28, by new contributor Steven wegmann of Cisco and ICSI).
  • Perceptual Audio Coding – MPEG audio and the related psychoacoustics (Chapter 35, by Dan Ellis).
  • Music Signal Analysis – automatic transcription of music (Chapter 37, by Dan Ellis).
  • Music Retrieval – music retrieval, including cover song detection (Chapter 38, by Dan Ellis).
  • Source Separation – methods to separate different signals, including CASA and multi-microphone methods (Chapter 39, by Dan Ellis, with a section on microphone arrays by Michael Seltzer of Microsoft Research).
  • Speaker Diarization – determining who spoke when (Chapter 42, by new contributor Gerald Friedland of ICSI).

Two other chapters have essentially been entirely rewritten: Speech Synthesis (Chapter 30, by Simon King from Edinburgh University), and Speaker Verification (Chapter 41, by David van Leeuwen from TNO). Also, Eric Fosler (of Ohio State University) has extensively revised his chapter on Linguistic Categories for Speech Recognition (Chapter 23).

Many other chapters have also undergone significant revisions; for instance, there are a number of significant updates to the chapters on ASR history (Chapter 4) and on feature extraction for ASR (Chapter 22), and a brief description of the Support Vector Machine (SVM) has been added to the deterministic pattern classification chapter (Chapter 8) in recognition of its increased importance. Finally, the Introduction has been modified to reflect the new distribution of chapters.

0.3 A FINAL THOUGHT

Ben Gold was the key inspiration and co-author for the first edition; there clearly would have been no book without him. He also was an inspiration and role model for me (Morgan) personally. It saddens me that he cannot be here for the new edition, but I know that his generous spirit would have welcomed the new contributions from Dan Ellis and others.

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset
35.172.230.21