Call for






Consistent & Reliable Acoustic Cues
for sound analysis


for a Speech Communication Special Issue

Recognition and Organization of Real-World Sound:
Papers related to the
2001 Workshop on Consistent and Reliable Acoustic Cues, CRAC-2001


Classification and recognition of acoustic signals--particularly speech--has reached a level of sophistication enabling many practical applications. The best recognition systems, however, make strong assumptions about the nature of the signal and any possible interference or noise.

At the same time, there is a growing body of research concerned with a more general class of audio signals, and dealing with them within a more flexible framework. From noise-robust speech recognition, through automatic transcription of recorded music, to various forms of source separation, this research acknowledges that getting information from sound is more than a simple pattern recognition problem--not least because even defining what information is desired is a challenge, and in almost every case the information must be extracted at many overlapping levels.

This special issue will represent a spectrum of work unified by the common theme of handling real-world sounds, with all the implicit complications. By covering a range of different processing strategies and a range of different acoustic analysis tasks, we hope to capture and showcase the variety, and the commonality, in these problems.

This special issue is a follow-on to a one-day meeting, the Workshop on Consistent and Reliable Acoustic Cues (CRAC-2001), held as a satellite event to Eurospeech 2001 in Aalborg, Denmark, last September. Although the common theme makes material presented at that workshop particularly appropriate, this call is completely open, and all relevant papers are equally welcome.


Submissions should describe previously unpublished research related to the following topics:

  • Identifying sources and extracting information in realistic acoustic environments, including:

      Auditory modeling
      Computational auditory scene analysis
      Blind source separation
      SNR estimation
      Confidence measures

  • Robust speech recognition strategies, including:

      Missing data techniques
      Noise modeling and compensation
      Multi-stream recognition

  • Applications to nonspeech acoustic sources such as musical instrument segregation and identification

  • Psychoacoustics of corrupt and noisy speech and audio perception

  • Combinations of any of the above, such as robust speech recognition based on computational auditory scene analysis


Full papers should be submitted to the special issue editors before

    May 15th 2002

The preferred format of submission is as a PDF file, via email to crac@ee.columbia.edu.

Papers are expected to be about 10 pages long in their final form; we request that double-spaced review versions be a maximum of 30 pages long.

We will conduct the first pass of reviewing over the summer, and hope to have the selected papers in their final form by September 2002, with the special issue to appear about 6 months later.


Special issue editors:

Updates/information will appear on the CRAC website,

This is a paragraph of dummy text whose purpose is to see if that by adding a line of text long enough to span more than the entire window we can in fact persuade the table rendering algorithms to respect the width=pixels specification of the leftmost column as an absolute maximum, rather than stretching it when the rightmost column is thinner.

last update: Sun Aug 18 23:11:28 EDT 2002 dpwe@ee.columbia.edu