Advances in Models for Acoustic Processing

Neural Information Processing Systems Workshop, Whistler, Canada, 9th Dec 2006

Click here to see the full programme of talks and extended abstracts

Description of this NIPS workshop

The analysis of audio signals is central to the scientific understanding of human hearing abilities as well as in engineering applications such as sound localisation, hearing aids or music information retrieval. Historically, the main mathematical tools are from signal processing: digital filtering theory, system identification and various transform methods such as Fourier techniques. In recent years, there is an increasing interest for Bayesian treatment and graphical models which permit more refined analysis and representation of the acoustic signals.

The application of Bayesian techniques is quite natural: acoustical time series can be conveniently modelled using hierarchical signal models by incorporating prior knowledge from various sources: from physics or studies of human cognition and perception. Once a realistic hierarchical model is constructed, many tasks such as coding, analysis, restoration, transcription, separation, identification or resynthesis can be formulated consistently as Bayesian posterior inference problems. In particular, the development of a powerful framework for approaching such tasks is central to improvements in the understanding of how both natural and synthetic systems may produce efficient auditory representations.

Goals of the Workshop

The goal of the workshop is to establish a discussion forum between practitioners of acoustical signal processing, researchers interested in computational neural acoustic processing, and more theoretically oriented researchers in machine learning, statistics and signal processing. This includes also researchers interested in the development of efficient neural codes for auditory representation. In particular, we welcome contributions that introduce interesting and challenging models for acoustical signal analysis and related inference techniques.

Example issues are:

  • What types of modelling approaches are useful for acoustic processing (e.g. hierarchical, generative, discriminative) ?
  • What classes of inference algorithms are suitable for these potentially large and hybrid models of sound ?
  • How can we improve the quality and speed of inference ?
  • Can efficient online algorithms be developed?
  • How can we learn efficient auditory codes based on independence assumptions about the generating processes?
  • What can biology and cognitive science can tell us about acoustic representations and processing?

The workshop scope is deliberately broad so that key advances in acoustic processing in natural systems, together with advances in computational modelling and inference methods may be discussed by experts that may not otherwise share the same platform. In so doing, we hope that a deeper understanding of acoustic processing, representations and applications may emerge.

From the responses of potential participants, we were able to identify three facets of acoustic processing :

  • FACET 1 (CNS) Computational Neuroscience/modelling of Auditory Organisation
  • FACET 2 (AUDIO) Models and inference techniques for Audio and Music Applications
  • FACET 3 (INF) Source Separation, Statistical Inference for analysis and processing of natural sounds

The workshop will be programmed to highlight the overlaps between these topics to provoke interaction. In addition to oral and poster presentations, two invited tutorials will also be given. Based on submissions, we will identify themes related to the workshop topics in order to group the talks and organize thematic discussions after the presentations.


David Barber IDIAP Research Institute

Taylan Cemgil University of Cambrige