Table 9 Recognition results on the evaluation set using trigram models in the first pass and context-dependent models in the second pass
FLMs | Optimal model order | WER [%] | Relative improvement[%] |
|---|---|---|---|
P | 3 | 25.90 | −1.01 |
M | 5 | 25.39 | 0.99 |
W | 5 | 25.46 | 0.70 |
P+M | 5 | 25.61 | 0.14 |
P+W | 5 | 25.95 | −1.20 |
M+W | 2 | 25.20 | 1.73 |
All | 5 | 25.45 | 0.74 |