By default, IDOL Speech Server automatically loads and unloads language packs depending on the requirements of server actions. For example, if you submit a French audio file to IDOL Speech Server for speech-to-text processing with the appropriate parameters, IDOL Speech Server automatically loads the French language pack.
Holding language packs in memory requires a lot of RAM. It also takes time to load language packs into memory. As a result, IDOL Speech Server actively manages the loading and unloading of language packs. The MaxLangResources
configuration parameter defines the maximum number of language packs that can be loaded in IDOL Speech Server simultaneously. For more information about this configuration parameter, see the IDOL Speech Server Reference.
You can override automatic loading and unloading of language packs and perform these tasks manually. Loading a language pack manually ensures that the model required for a task is ready for use, which reduces startup time for tasks that you add later. Manually loaded language packs remain loaded until you manually unload them.
Note: The exact task names and action parameters to use depend on the configuration in the IDOL Speech Server tasks configuration file.
To manually load a language pack
Send a LoadLanguage
action with the Name
parameter set to the name of the configuration file section where the language pack that you want to load is defined.
If you want to add new words to the language model, or increase the weighting of specific words, you can specify a text file that contains the relevant words and weights as the value of the ClassWordFile
parameter. You can set this parameter for any Speech Server task that uses language resources.
For example:
http://localhost:13000/action=LoadLanguage&Name=ENUS&ClassWordFile=D:/WordWeightings.txt
This action uses port 13000
to instruct IDOL Speech Server, which is located on the local machine, to load the ENUS
language to IDOL Speech Server, using the additional words and weightings specified in the WordWeightings.txt
class word file.
You can also add new pronunciations or edit existing pronunciations at language load time. To do this, specify a text file that contains the pronunciations that you want to add or edit as the value of the PronFile
parameter. You can set this parameter for any Speech Server task that uses language resources.
For example:
http://localhost:13000/action=LoadLanguage&Name=ENUK-6.3&PackDir=ENUK&PronFile=T:\Pronunciations.txt
This action uses port 13000
to instruct IDOL Speech Server, which is located on the local machine, to use Pronunciations.txt
to update the pronunciations for the ENUK-6.3
language pack in the ENUK
directory.
For more information on the format of the pronunciations file, see the IDOL Speech Server Reference.
Note: If you load a language pack with a class word file or a pronunciations file, you cannot add new entries to the file by editing the file.To add new entries, you must create a new file in the same format and rerun the action. When you do this, Speech Server loads a new language pack. Note that this applies to automatically loaded language packs as well as manually loaded language packs. For more information, see the IDOL Speech Server Reference.
To manually load a custom language model
Send a LoadLanguage
action:
Name
parameter to the name of the language pack section in the configuration file.DctFile
parameter to the name of the custom dictionary file.CustomLM
parameter.Specify a list of .tlm language files and weights separated by colons (:), in the form file:weight
.
For example, if you use two files, autn1.tlm
and autn2.tlm
, and you want to set the weights to 0.3
and 0.4
respectively, use:
CustomLM=autn1:0.3:autn2:0.4
For example:
http://localhost:13000/action=LoadLanguage&Name=ENUS&CustomLM=hpe1:0.3:hpe2:0.4&DctFile=hpe
This action uses port 13000
to instruct IDOL Speech Server, which is located on the local machine, to load the ENUS
base language model, the hpe1.tlm
custom language model, and the hpe2.tlm
custom language model. The models are weighted using the ratio 0.3
:0.3
:0.4
. IDOL Speech Server also loads the hpe
dictionary file.
To determine the weight of the base language model, subtract the weights of the custom language models from 1.0. In this case, the base language model has a weight of 0.3
. All weights must be greater than 0
and less than 1.0
.
|