Name
Abstract | ||
|---|---|---|
Within the framework of speaker-adaptation, a tech- nique based on tree structure and the maximum a poste- riori criterion was proposed (SMAP). In SMAP, the pa- rameters estimation, at each node in the tree is based on the assumption that the mismatch between the training and adaptation data is a Gaussian PDF which parameters are estimated by using the Maximum Likelihood crite- rion. To avoid poor transformation parameters estimation accuracy due to an insufcienc y of adaptation data in a node, we propose a new technique based on the maxi- mum a posteriori approach and PDF Gaussians Merging. The basic idea behind this new technique is to estimate an afne transformations which bring the training acous- tic models as close as possible to the test acoustic models rather than transformation maximizing the likelihood of the adaptation data. In this manner, even with very small amount of adaptation data, the parameters transforma- tions are accurately estimated for means and variances. This adaptation strategy has shown a signicant perfor- mance improvement in a large vocabulary speech recog- nition task, alone and combined with the MLLR adapta- tion. |
Year | Venue | Keywords |
|---|---|---|
2003 | INTERSPEECH | parameter estimation,tree structure,linear model,maximum likelihood |
Field | DocType | Citations |
Pattern recognition,Linear model,Computer science,Speech recognition,Gaussian,Tree structure,Artificial intelligence,Maximum a posteriori estimation,Maximum likelihood sequence estimation,Merge (version control),Speaker adaptation,Performance improvement | Conference | 1 |
PageRank | References | Authors |
0.35 | 12 | 5 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Driss Matrouf | 1 | 404 | 41.80 |
| Olivier Bellot | 2 | 8 | 1.59 |
| Pascal Nocera | 3 | 70 | 10.86 |
| Georges Linares | 4 | 87 | 19.73 |
| Jean-François Bonastre | 5 | 64 | 10.60 |