ESPRESSO II:
Continuous-state dynamical system models for speech recognition

The second grant for the Espresso project ran from 1st October 1999 to 30th Sept 2000 and 1st April 2001 to 31st Jan 2002.

Introduction

In this, the second phase of Espresso, we investigated new acoustic models which account for the continuous, asynchronous nature of speech. We chose to look at linear, continuous-state dynamical system models, also known as Kalman filters.

Personnel

• Simon King (PI)
• Steve Isard (PI)
• Also associated with the project were two Ph.D. students:
  • Korin Richmond (acoustic-to-articulatory mapping)
  • Joe Frankel (Monte-Carlo methods for training linear dynamical systems)

Publications

• You can find all the publications relating to Espresso here.
• Here are some slides from talks on Espresso II
  • IoA 2000: Frankel
  • IoA 2000: King, Taylor
  • IoA 2000: Richmond
  • March 2000
  • PhonASR 2000
• The final report [ps|pdf] to the EPSRC is also available.

Funding

This work is being funded by the Engineering and Physical Science Research Council, EPSRC grant number GR/M75204/01 "Continuous-state dynamical system models for speech recognition" under the "Realising Our Potential Awards" (ROPA) scheme.

Contact Simon King for more details.