IBM was one of the first companies to produce a commercial application built upon speech recognition (also aimed at the telephony market), and ViaVoice is their latest speech recognition system. ViaVoice runs in one of two modes; Dictation Mode or Command and Control (C&C) Mode. The first uses a large database of words and although allowing targeted lexicons for specific domains (for example medical or financial institutions) there is little control over the utterances permitted by the application. The C&C mode however, requires each application to supply its own grammar for the language the program will accept.
The engine is supplied with a large amount of pronunciation details, but if a new word that has no pronunciation stored in the engine's databases is included within an applications grammar a pronunciation must be given to the engine as well.
The grammars supplied to ViaVoice in C&C mode conform to the SRCL standard (see Section 2.1.2).
It should be noted that both these modes fall into the continuous speech recognition category, and no commercial tool will sell well if it uses isolated word recognition. Of course using C&C mode with a very limited grammar, such as one terminal per pre-terminal and only one level of rule expansion, effectively places the mode into isolated word recognition, but this is the application designers choice, not a limitation of the recognition engine.
An example of a system which uses ViaVoice is ABSTRACTOR [31] which uses SR and natural language understanding in order to add abstracts to documents.