“2013-08-09”版本间的差异
来自cslt Wiki
（以内容“== Data sharing == * LM count files still undelivered! == DNN progress == === Experiments === * Discriminative DNN Use sequential MPE/MMI/bMMI(0.1) (with the DNN-b...”创建新页面） |
（→DNN progress） |
||
| 第5行： | 第5行： | ||
== DNN progress == | == DNN progress == | ||
| − | === | + | === Discriminative DNN === |
| − | * | + | * Running 1200-3620 NN, 1000 hour MMI training. 2% absolute improvement is expected. |
| − | + | === Sparse DNN === | |
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| + | * Iterative sparse sticky training runs. More sparsity is expected. | ||
=== Tencent exps === | === Tencent exps === | ||
| − | |||
| − | |||
| − | |||
| − | |||
==Confidence estimation== | ==Confidence estimation== | ||
2013年8月9日 (五) 08:33的版本
目录
Data sharing
- LM count files still undelivered!
DNN progress
Discriminative DNN
- Running 1200-3620 NN, 1000 hour MMI training. 2% absolute improvement is expected.
Sparse DNN
- Iterative sparse sticky training runs. More sparsity is expected.
Tencent exps
Confidence estimation
DNN confidence
- Waiting for Puqiang's results.
- Puqiang reported a bug, waiting for debug.
- To be done:
- CI phone state confidence
Multi-graph decoding based on DNN confidence
- Lenovo keep testing the DNN-confidence-based multi graph decoding.
Subgraph intergration
- G.fst integration is done. Initial test passed.
- HCLG integration is done. Waiting for test.
Embedded progress
- ARM so building completed
- 1000 words RT 0.8, 2300 RT with small increasement.
- To be done
- discriminative training for small network.
