Abstract | ||
---|---|---|
We consider a family of Gaussian mixture models for use in HMM based speech recognition system. These "SPAM" models have state independent choices of subspaces to which the precision (inverse covariance) matrices and means are restricted to belong. They provide a flexible tool for robust, compact, and fast acoustic modeling. The focus of this paper is on the case where the means are unconstrained. The models in the case already generalize the recently introduced EMLLT models, which themselves interpolate between MLLT and full covariance models. We describe an algo- rithm to train both the state-dependent and state-independent pa- rameters. Results are reported on one speech recognition task. The SPAM models are seen to yield significant improvements in accu- racy over EMLLT models with comparable model size and run- time speed. We find a relative reduction in error rate over an MLLT model can be obtained while decreasing the acoustic mod- eling time by . |
Year | Venue | Keywords |
---|---|---|
2002 | INTERSPEECH | gaussian mixture model,speech recognition,error rate,comparative modeling |
Field | DocType | Citations |
Covariance function,Estimation of covariance matrices,Subspace topology,Pattern recognition,Matrix (mathematics),Computer science,Law of total covariance,Covariance intersection,Artificial intelligence,Covariance mapping,Covariance | Conference | 35 |
PageRank | References | Authors |
2.32 | 5 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Scott Axelrod | 1 | 113 | 10.14 |
Ramesh Gopinath | 2 | 123 | 10.65 |
Peder A. Olsen | 3 | 398 | 37.80 |