Standard and Custom Language Models

Before you run a speech processing operation, you must decide whether to use a standard language model or a custom model. This table summarizes the main features of both types of language model.

Standard language model Custom language model
Available out of the box Need to create yourself using IDOL Speech Server tools
Even coverage of topics Focuses on particular topics, which increases accuracy
Trained on billions of words Generally trained on a smaller number of words; however, you can combine it with the standard language model when you perform tasks, to increase the coverage
Might not cover specialist vocabulary Can cover specialist vocabulary, such as technical terms or product names

You can also estimate the suitability of a language model for a task by calculating its perplexity (see Calculate Perplexity).


_HP_HTML5_bannerTitle.htm