Digital Audio Restoration: A Statistical Model Based Approach

Rating:
Summary: the
Review: This book was written by two professors at Cambridge University, who are also founders of the CEDAR Company, which is on the leading edge of audio restoration technology. The book consisely summarizes several of the audio restoration ideas developed by the authors that have been published in various journals. The book focuses on Bayesian analysis as a means for determining the most probable nature of audio defects (clicks, pitch variations and so forth) based on observed, corrupted data. Bayesian inference is also used to find maximum a posteriori characteristics of the underlying, uncorrupted signal.

Those interested in restoring old phonograph and cylinder recordings will find the book very interesting, should one want to understand and develop audio restoration software.

The book is very well-written and typeset, unlike many contemporary technical books. It uses vector and matrix notation for most formulae, which is welcome to those familiar with matlab. There is a very good list of references.

Many of the priniciples discussed require a good knowledge of statistical signal processing by the reader. (Fortunately a new book "Computational Statistics Handbook with Matlab" was just published and should fill some gaps.)

The book comes with an audio CDROM that illustrates several of the techniques discussed in the book. The CD itself is effectively an advertisement for their CEDAR sound processing systems, as the CD has clips with varying levels of processing applied, including the full CEDAR proprietary system. The CD contains the original source material as wav files, so researchers can use them to benchmark their algorithms against the commercial CEDAR system.

I find it disappointing that the authors did not include matlab code to better illustrate their concepts. But then, the authors caution that some of the algorithmic details embodied in their restoration software are not published "for reasons of commercial secrecy."

[As an interesting side-note, one other founder of CEDAR, S. Vaseghi, has published a book on noise reduction and audio restoration. The publisher's advertisement said it came with matlab code. This is not the case. I contacted the publisher and was told (three months later) that the matlab reference was an oversight. Why the secrecy?]

In summary, Digital Audio Restoration is an excellent book by two of the world's leading researchers in this area. I would have given it a five-star rating if it had matlab (or C) reference code. An intrepid researcher could sift through IEEE Transactions on Signal Processing and ICASSP proceedings and glean much of what's in this book; thus matlab examples would have been a welcome addition. Still, there's no better book to my knowledge in the area of audio restoration.

Rating:
Summary: the
Review: This book was written by two professors at Cambridge University, who are also founders of the CEDAR Company, which is on the leading edge of audio restoration technology. The book consisely summarizes several of the audio restoration ideas developed by the authors that have been published in various journals. The book focuses on Bayesian analysis as a means for determining the most probable nature of audio defects (clicks, pitch variations and so forth) based on observed, corrupted data. Bayesian inference is also used to find maximum a posteriori characteristics of the underlying, uncorrupted signal.

Those interested in restoring old phonograph and cylinder recordings will find the book very interesting, should one want to understand and develop audio restoration software.

The book is very well-written and typeset, unlike many contemporary technical books. It uses vector and matrix notation for most formulae, which is welcome to those familiar with matlab. There is a very good list of references.

Many of the priniciples discussed require a good knowledge of statistical signal processing by the reader. (Fortunately a new book "Computational Statistics Handbook with Matlab" was just published and should fill some gaps.)

The book comes with an audio CDROM that illustrates several of the techniques discussed in the book. The CD itself is effectively an advertisement for their CEDAR sound processing systems, as the CD has clips with varying levels of processing applied, including the full CEDAR proprietary system. The CD contains the original source material as wav files, so researchers can use them to benchmark their algorithms against the commercial CEDAR system.

I find it disappointing that the authors did not include matlab code to better illustrate their concepts. But then, the authors caution that some of the algorithmic details embodied in their restoration software are not published "for reasons of commercial secrecy."

[As an interesting side-note, one other founder of CEDAR, S. Vaseghi, has published a book on noise reduction and audio restoration. The publisher's advertisement said it came with matlab code. This is not the case. I contacted the publisher and was told (three months later) that the matlab reference was an oversight. Why the secrecy?]

In summary, Digital Audio Restoration is an excellent book by two of the world's leading researchers in this area. I would have given it a five-star rating if it had matlab (or C) reference code. An intrepid researcher could sift through IEEE Transactions on Signal Processing and ICASSP proceedings and glean much of what's in this book; thus matlab examples would have been a welcome addition. Still, there's no better book to my knowledge in the area of audio restoration.