Name
Abstract | ||
|---|---|---|
This paper proposes a new covariance modeling technique for Gaussian Mixture Models. Specifically the inverse covariance (precision) matrix of each Gaussian is expanded in a rank-1 basis i.e., Σj−1 = Pj = Σk = 1D λkjakakT, λkj ∈ ℝd. A generalized EM algorithm is proposed to obtain maximum likelihood parameter estimates for the basis set {akakT} and the expansion coefficients {λkj}. This model, called the Extended Maximum Likelihood Linear Transform (EMLLT) model, is extremely flexible: by varying the number of basis elements from d to d(d + 1)/2 one gradually moves from a Maximum Likelihood Linear Transform (MLLT) model to a full-covariance model. Experimental results on two speech recognition tasks show that the EMLLT model can give relative gains of up to 35% in the word error rate over a standard diagonal covariance model. |
Year | DOI | Venue |
|---|---|---|
2002 | 10.1109/ICASSP.2002.5743949 | ICASSP), 2002 IEEE International Conference |
Keywords | Field | DocType |
covariance modeling technique,emllt model,generalized em algorithm,speech recognition,maximum likelihood estimation,gaussian mixture models,covariance matrices,extended maximum likelihood linear transform model,maximum likelihood parameter estimates,matrix inversion,inverse covariance matrices,density functions,basis expansion,gaussian mixture model,linear transformation,word error rate,em algorithm,indexing terms,maximum likelihood | Inverse,Pattern recognition,Matrix (mathematics),Gaussian,Artificial intelligence,Sigma,Covariance matrix,Estimation theory,Mathematics,Covariance,Lambda | Conference |
Volume | Issue | ISSN |
1 | 1 | 1520-6149 |
ISBN | Citations | PageRank |
0-7803-7402-9 | 32 | 2.55 |
References | Authors | |
10 | 2 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| P. Olsen | 1 | 85 | 10.62 |
| Ramesh Gopinath | 2 | 123 | 10.65 |