Weight of language model during rescore phase. Absolute value (not dependent on swirec_lmweight).
|
Value |
Integer: 0–INT_MAX. |
|
Default |
0 |
|
How to set |
Set in the browser, using the Vendor-Specific-Parameters header in a RECOGNIZE or SET-PARAMS method. |
|
Usage |
Rarely used |
Depending on the power of the language model to predict utterances, you can tune the language-model weight up or down in order to improve accuracy. However, if the language model weight is tuned up too far, it routinely discards unlikely hypotheses before it receives a lot of evidence. One way to resolve this is to keep swirec_lmweight at a “middle” value and to increase swirec_lmrescore_weight. Since swirec_lmrescore_weight is applied after the full hypothesis evaluation, it is less sensitive.
Related parameter
