3-state monofeat model with L, T, G connected to dg1, pl1, etc.
<3-state-monofeat-cloud>
| | |
| | |
V V V
L _____T G
|\ / \ /|
| /\_______\___/ |
|/_/\______ \___|
V \____ V
dg1 pl1 ...
| |
| |
V V
VE_dg1 VE_pl1
where dg1, pl1, etc. depend on L, T, G.
The relationship between each of the features and L, T, G can be non-deterministic (a dense CPT) or non-deterministic (implemented as a dense CPTs with only one non-zero per row).
| Models trained use original work alignments |
| Vocab size | Task | Word error rate | Language model | Det CPTs? | Notes |
| Validation | Test | Scale | Penalty |
| 10 | 1 | 43.8 | | 36 | -2 | no | cbeam=200 |
Note: a small number of utterances must be re-run with a smaller beam.
--
SimonKing - 14 Aug 2006
