Using a custom language model can substantially improve speech-to-text rates, if the model represents the speech data very accurately.
For instructions on how to build custom language models, see Create Custom Language Models. After you build a custom language model, you must add it to the language pack section in the tasks configuration file (see Configure Language Packs).
To use a custom language model for speech-to-text
CustomLm
parameter when you send the AddTask
action. Set it to the name of the custom language model and the interpolation weight, separated by a colon (:). For example:
http://localhost:13000/action=AddTask&Type=SpeechToText&File=C:/myData/Speech.wav&Out=SpeechTranscript.ctm&Lang=ENUS&CustomLm=myLangModel:0.4
This action uses port 13000
to instruct IDOL Speech Server, which is located on the local machine, to perform the SpeechToText
task on the Speech.wav
file using both the U.S. English and myLangModel
language packs, and to write the results to the SpeechTranscript.ctm
file. The myLangModel
language pack is weighted at 0.4
against the U.S. English pack.
IDOL Speech Server suggests the interpolation weight to use at the time that you build the custom language model. All custom language models are placed in a specific folder, so you need to specify only the base name of the custom language model.
|