Publication Details
Title: Scaling Up: Learning Large-scale Recognition Methods from Small-scale Recognition Tasks
Author: N. Morgan, B. Y. Chen, Q. Zhu, and A. Stolcke
Group: ICSI Technical Reports
Date: September 2003
PDF: ftp://ftp.icsi.berkeley.edu/pub/techreports/2003/tr-03-002.pdf
Overview:
Despite the common wisdom that lessons learned from small experimental speech recognition tasks often do not scale to larger tasks, many important algorithms used in larger tasks were first developed with small systems applied to small tasks. In this paper we report experiments with the OGI Numbers task that led to the adoption of a number of engineering decisions for the design of an acoustic front end. We then describe a three-stage process of scaling to the larger conversational telephone speech (CTS) task. Much of the front end design required no change at all for the more difficult task, yielding significant improvements over our baseline front end.
Bibliographic Information:
ICSI Technical Report TR-03-002
Bibliographic Reference:
N. Morgan, B. Y. Chen, Q. Zhu, and A. Stolcke. Scaling Up: Learning Large-scale Recognition Methods from Small-scale Recognition Tasks. ICSI Technical Report TR-03-002, September 2003
Author: N. Morgan, B. Y. Chen, Q. Zhu, and A. Stolcke
Group: ICSI Technical Reports
Date: September 2003
PDF: ftp://ftp.icsi.berkeley.edu/pub/techreports/2003/tr-03-002.pdf
Overview:
Despite the common wisdom that lessons learned from small experimental speech recognition tasks often do not scale to larger tasks, many important algorithms used in larger tasks were first developed with small systems applied to small tasks. In this paper we report experiments with the OGI Numbers task that led to the adoption of a number of engineering decisions for the design of an acoustic front end. We then describe a three-stage process of scaling to the larger conversational telephone speech (CTS) task. Much of the front end design required no change at all for the more difficult task, yielding significant improvements over our baseline front end.
Bibliographic Information:
ICSI Technical Report TR-03-002
Bibliographic Reference:
N. Morgan, B. Y. Chen, Q. Zhu, and A. Stolcke. Scaling Up: Learning Large-scale Recognition Methods from Small-scale Recognition Tasks. ICSI Technical Report TR-03-002, September 2003