Speech Communication: Special interest article

The Speech Communication webpages feature full-text articles of particular interest for the readership. Together with the Editor-in-Chief of the journals we will regularly select new articles for display in this section.

The papers presented here are in PDF format, so if you have not yet installed your Acrobat(R) reader yet, download a free copy of this helper application from Adobe(R)

Towards increasing speech recognition error rates

H. Bourlard[a, b], H. Hermansky[a, d] and N. Morgan[a, c]

[a] International Computer Science Institute, Berkeley, CA, U.S.A.
[b] Faculté Polytechnique de Mons, Mons, Belgium
[c] University of California at Berkeley, Berkeley, CA, U.S.A.
[d] Oregon Graduate Institute, Portland, OR, U.S.A.

Summary
This article is intended for all that have an interest in Automatic Speech Recognition (ASR). It argues that in order to make significant advances in ASR one may need to momentarily accept some decrease in the performance of existing systems. The paper also sketches some innovative research directions that provide some promise for the future.

Note
This article corresponds to an extended and revised version of the keynote presented by the first author at EUROSPEECH '95 in Madrid. The help of R. De Mori, F. Jelinek, R. Lippmann, J. Makhoul and V. Steinbiss who agreed to review the first version of this article is greatly acknowledged by the editors of Speech Communication.

The Editor-in-Chief, Christel Sorin, has invited leading experts in this field to comment on the thesis that is put forward in this article. In the same issue 18(3) as well as on this Webserver the comments are published of:

  1. B. Atal;
  2. R. De Mori;
  3. J. Flanagan;
  4. S. Furui;
  5. J.P. Haton;
  6. M.J. Hunt
  7. F. Jelinek;
  8. R. Lippmann;
  9. J. Mariani, J.L. Gauvain and L.Lamel; and
  10. S. Seneff.
The editors hope that the readers of Speech Communication will benefit and enjoy this stimulating and lively issue of the journal as well as its public availability on the World Wide Web, and they thank again warmly all its contributors.