The ContextSize
parameter sets the group size to use when combining features during training.
The default value of 3 is generally suitable for most data. However, if you have poor quality audio data, a lower value can improve the results, because it is more robust to errors in the underlying feature process. In general, HPE recommends that you use a value of 3 for broadband data, and 2 for telephony.
*You must set the value of this configuration parameter by setting the corresponding action parameter.
Type: | Integer |
---|---|
Default: | 3 |
Allowed Range: | Minimum: 1 Maximum: 5 |
Required: | No |
Configuration Section: | lidtrain module |
Example: | ContextSize=4
|
See also: | Context (action parameter) |
|