Some ways to classify a speech recognition system and speech quality have been described. Generally, a recognition system can only cope with some of the parameters in the ``difficult'' region, and must yield towards the other conditions. For instance, a large vocabulary speaker independent continuous speech recogniser only performs well on undisturbed, read speech of a known domain (vocabulary and grammar ), and a speaker-independent telephone speech real-time recognition system may be designed to recognise digits only. In this way, the capabilities of a particular recogniser with respect to all the conditions define a capability profile.
It is a very good idea to realise that the requirements of a
particular application define a similar application
requirement profile which is generally different from the
capability profile of most recognition systems.
RECOMMENDATION 1
Before you choose your set of recognisers to assess, define the
requirements of the application, if any, and compare those with the
specifications of the recognition systems.