Loading...
Please wait, while we are loading the content...
Similar Documents
Linear Feature Space Projections For Speaker Adaptation (2001)
| Content Provider | CiteSeerX |
|---|---|
| Author | Padmanabhan, Mukund Zweig, Geoffrey Saon, George |
| Description | Proceedings of ICASSP 2001 |
| Abstract | We extend the well-known technique of constrained Maximum Likelihood Linear Regression (MLLR) to compute a projection (instead of a full rank transformation) on the feature vectors of the adaptation data. We model the projected features with phone-dependent Gaussian distributions and also model the complement of the projected space with a single class-independent, speaker-specific Gaussian distribution. Subsequently, we compute the projection and its complement using maximum likelihood techniques. The resulting ML transformation is shown to be equivalent to performing a speaker-dependent heteroscedastic discriminant (or HDA) projection. Our method is in contrast to traditional approaches which use a single speaker-independent projection, and do speaker adaptation in the resulting subspace. Experimental results on Switchboard show a 3% relative improvement in the word error rate over constrained MLLR in the projected subspace only. |
| File Format | |
| Publisher Date | 2001-01-01 |
| Access Restriction | Open |
| Subject Keyword | Single Speaker-independent Projection Single Class-independent Full Rank Transformation Well-known Technique Adaptation Data Linear Feature Space Projection Feature Vector Projected Space Speaker Adaptation Constrained Maximum Likelihood Linear Regression Speaker-specific Gaussian Distribution Traditional Approach Phone-dependent Gaussian Distribution Word Error Rate Ml Transformation Projected Feature Speaker-dependent Heteroscedastic Discriminant Relative Improvement Resulting Subspace Projected Subspace Experimental Result Maximum Likelihood Technique |
| Content Type | Text |
| Resource Type | Proceeding |
