Release 3 of Dragon Systems Inc’s Naturally Speaking product looks likely to add more heat to the ‘next big thing’ atmosphere surrounding voice recognition. However, despite the falling price and high accuracy rate of this and other new speech recognition packages, voice recognition systems still have a few years to go before they become widely […]
Release 3 of Dragon Systems Inc’s Naturally Speaking product looks likely to add more heat to the ‘next big thing’ atmosphere surrounding voice recognition. However, despite the falling price and high accuracy rate of this and other new speech recognition packages, voice recognition systems still have a few years to go before they become widely accepted. Certainly, the new system from the Boston, Massachusetts-based company – released in three different versions – Standard, Preferred and Professional, and a ‘no frills’ consumer edition, is faster and more accurate than previous editions. Along with new packages from IBM Corp (CI No 3,427) and Philips Electronic NV (CI No 3,422), NaturallySpeaking is pushing the take-up of voice recognition software in the home and small office. To this end, Dragon has recently signed a bundling deal with Corel Corp. NaturallySpeaking will be included as part of WordPerfect 8 suite and IBM will integrate ViaVoice into Lotus Development Corp’s SmartSuite. Only Microsoft Corp’s Office 2000 will be bereft of a speech recognition element, although all of the other packages are compatible with Microsoft Word.
However, the main obstacle for all of the new applications is the high demand they place on hardware. The Dragon Naturally Speaking Preferred Edition system recommendation is a 200MHz Pentium and a minimum of 48MB RAM – the application takes up a hefty portion of your system – if you have a system capable of running it, many offices and homes do not. Even if your system is capable of running NaturallySpeaking, do you want an application that makes such a demand on your CPU? This lag in processing power is what Dragon describes as its only real problem. A solution would be a dedicated chip or board with onboard memory, similar in concept to 3D graphics boards currently available. Any such board would implement speech recognition routines – which are currently implemented in software – in hardware thus speeding up the routines and freeing up valuable processor space. Now, 2D/3D graphics boards are increasingly issued as standard on retail PCs – and you would not buy a PC without a sound card or multimedia card – a cheap, dedicated speech recognition card could become just as de rigeur. Dragon, which claims to have a real close connection with Intel Corp – having collaborated with the chip giant on the MMX extensions for the Pentium series – would seem well placed to collaborate again with Intel on a card for speech recognition purposes. A low-cost voice chip would facilitate the production of mobile translation devices and speech-driven personal digital assistants; a market that Lernout & Hauspie Speech Product NV seems to be increasingly interested in (CI No 3,368). Dragon has already included a mobile element on its new package where users can dictate speech onto a Sony Corp MiniDisc recorder (MZ-R30, MZ-R50 or MZ-B3), or a Norcom 2500 (with speech recognition coupler) mini-cassette recorder. This can then be plugged into a PC and the application will create and transcribe a file from the recording.