Perplexities for each RNN LM and test-set. Note that MS was only used for training the LM, whereas JNAS speech was used for evaluating the performance of the MS LM, as described in Section 5.1.1
| LM | Test-set | ||||
|---|---|---|---|---|---|
| APS | SPS | CV | CEJC | JNAS | |
| APS | 17.3 | 33.4 | 93.3 | 43.1 | 147.7 |
| SPS | 34.8 | 17.4 | 56.9 | 32.1 | 100.6 |
| CV | 179.0 | 92.8 | 17.4 | 86.5 | 195.6 |
| CEJC | 99.2 | 49.5 | 88.4 | 23.8 | 208.3 |
| MS | 95.8 | 56.6 | 46.6 | 50.8 | 33.4 |
| Test-set | |||||
|---|---|---|---|---|---|
| 17.3 | 33.4 | 93.3 | 43.1 | 147.7 | |
| 34.8 | 17.4 | 56.9 | 32.1 | 100.6 | |
| 179.0 | 92.8 | 17.4 | 86.5 | 195.6 | |
| 99.2 | 49.5 | 88.4 | 23.8 | 208.3 | |
| 95.8 | 56.6 | 46.6 | 50.8 | 33.4 | |
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.