Experiments with Grand Variance in the Arm Continuous Speech Recognition System
Abstract
The use of triphones to cope with contextual effects in phoneme-level hidden Markov model (HMM) based speech recognition results in a huge increase in the number of system parameters which need to be estimated. The solution to this problem is to reduce the number of independent system parameters so that those which remain can be estimated more robustly from the training data. For HMMs with Gaussian state output probability density functions (pdfs), a simple example of such an approach is the 'grand' variance method in which all state output pdfs share the same covariance matrix. This paper reports the results of experiments designed to investigate the effect of grand variance on the performance of the triphone-HMM based ARM continuous speech recognition system. (JHD)
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 08, 1990
- Accession Number
- ADA221728
Entities
People
- K. M. Ponting
- M. J. Russell
Organizations
- Royal Signals and Radar Establishment