Dictation

Dragon Systems' Naturally Speaking

IBM's ViaVoice

Lernout & Hauspie's Voice Xpress

Philips' FreeSpeech.

However, there is also considerable scope for products which embed these or other products in an environment designed to favour effective generation of written text. One such example is the   CyberTranscriber system,  based on the ARMADA technology originated by DERA-SRU, and available over the Internet.

Other companies, as Entropic  (now absorbed by Microsoft), provide environments to customize speech products for particular purposes, providing access to the latest technology but without making a commitment to specific decisions on the best way to integrate it with business and other practices.

• Although a feature of many early spoken language recognition systems was simultaneous display of the text as it was dictated, it may be better, functionally, not to allow simultaneous editing, or indeed display. One of the major advantages of this technology is to speed up processing times, and the temptation to correct as one goes tends to slow throughput.

• The technology that has proved successful depends on a combination of Hidden Markov Models, to recognize sounds from their acoustics, and statistical language models, to estimate the likelihood of particular words in a given context. Both of these can only be built effectively with the aid of large amounts of background data, of annotated recordings and relevant text respectively. Without these in the background, it cannot be expected that CSR can be developed for a new language: but if these are available, the techniques used should be effective, whatever the language.