v7.1, 20040924, me

Friday, October 29, 9:00 am – 11:00 am
Session F AUDIO ARCHIVING, STORAGE, AND RESTORATION; CONTENT MANAGEMENT

Chair: David Ackerman, Harvard University, Cambridge, MA, USA

9:00 am
F-1 Wow Detection and Compensation Employing Spectral Processing of Audio—Andrzej Czyzewski, Przemyslaw Maziewski, Marek Dziubinski, Andrzej Kaczmarek, Bozena Kostek, Gdansk University of Technology, Gdansk, Poland
The engineered algorithms are presented for the detection of parasitic frequency modulation in audio originating from irregularities of sound carrier velocity. The algorithms were developed with special regard to non-periodic frequency modulation effects found in old movie sound tracks. The proposed algorithms consider the influence of the wow disturbance on the location of formants in time-frequency representation. The dynamic analysis of formant structure behavior underlies discriminating between parasitic frequency changes and natural frequency fluctuations. The compensation of the detected wow-related frequency modulation is accomplished basing on the non-uniform resampling algorithm, driven by the discerned parasite modulation patterns. The details of the proposed wow detection and compensation techniques are presented and achieved results are discussed.
Convention Paper 6212

9:30 am
F-2 Correction of Wow and Flutter Effects in Analog Tape Transfers—Jamie Howarth, Plangent Processes, Nantucket, MA, USA; Patrick Wolfe, University of Cambridge, Cambridge, UK
In this paper we describe a system whereby analog hardware is combined with the theory of nonuniform sampling in order to correct for wow and flutter effects in analog tape transfers. We show how in certain instances the medium itself can provide an accurate measurement of a recording’s timing irregularities, in which case digital signal processing techniques permit a playback-rate correction of what is essentially an irregularly sampled audio waveform. Results using both real and synthetic data demonstrate the effectiveness of the method, both in cases of severe degradation as well as high-quality analog transfers heretofore considered normal.
Convention Paper 6213

10:00 am
F-3 Computationally Efficient Blind Dereverberation of Audio Signals—Hesu Huang, Chris Kyriakakis, University of Southern California, Los Angeles, CA, USA
Convolutive noise in the form of reverberation can significantly degrade the quality and intelligibility of the real-world audio recordings. To reduce this type of acoustic noise, we propose a single-microphone dereverberation method based on Constant Modulus Algorithm (CMA)—a blind deconvolution technique. In particular, a new Noncausal Delayless Subband Filtering architecture is designed and combined with CMA to reduce the computational complexity of the overall system. Experimental results show that our method presents a comparable performance to full-band CMA but with less computational cost in dereverberating audio signals.
Convention Paper 6214

10:30 am
F-4 Audio Fingerprints: Technology and Applications— Douglas Keislar, Erling Wold, Thom Blum, Muscle Fish, a division of Audible Magic Corporation, Berkeley, CA, USA
Audio fingerprinting provides the base technology for a variety of recent applications. Aspects of the underlying fingerprinting technology and some typical applications are presented. An effort is made to discuss both Audible Magic’s work as well as that of its competitors.
Convention Paper 6215