IBM Corp has renamed its Personal Dictation System and it is now VoiceType Dictation, and it plans to introduce a selection of specialist vocabularies for the product, as well as a Windows version and a signal processor adaptor board for portable computers. IBM says users can dictate at speeds of up to 100 words a minute, at 97% accuracy with VoiceType. But users must speak distinctly, separating each word: although the system can cope with a certain level of mispronunciation or homophones, it cannot handle ordinary speech. Users can dictate without looking at the machine or immediately having to correct errors, features that IBM says makes its speech recognition product superior to others available now. Users can listen to what they have dictated, eliminating the need to remember what was said, so making correcting much easier. But it gobbles up memory: one minute of dictation uses 1Mb. The US-English version has a vocabulary of 32,000 words, the UK-English version, 29,500 words. Of the specialist vocabularies, a module for radiologists will be the first available. It was developed with Belfast City Hospital’s radiology department and has a vocabulary of 20,000 words, 6,000 of which are specific to the activity. It also understands the context of words, so colon becomes a part of the body rather than a punctuation mark. Other specialist language models planned are a legal one, already in beta test, and general medicine and journalism. However, when the specialist models are used they replace the standard ones. At VoiceType’s heart is a statistical language model that recognises the individual user’s voice and way of speaking: the more the user uses VoiceType, the more accurate it becomes. IBM reckons it takes between 45 minutes to an hour for VoiceType to get used to a new user. IBM provides 150 sentences to read to the machine, which it says is enough to build a voice model that will start with 95% accuracy. This specificity of the product is key but it means a radiologist using the radiology module on her lap top, who fancied dictating a romantic novel while travelling to work would find the specialist model inappropriate. To use VoiceType the user must switch the microphone on manually, whereafter everything is verbally driven including compatible applications loaded on the computer – Notes, Ami Pro, cc:Mail, Excel, Word, and Quicken.
Dictation window
The program transcribes the speech into a dictation window and not directly into an application. IBM says this was deliberate: by using the dictation window the company says the user is given more functionality. But, having said that, it is making application programming interfaces, hooks and tools available to developers via the Developers’ Assistance Programme, launched in October, so that they can integrate VoiceType into their applications. IBM says VoiceType is not something that can simply be added on to an application. Most existing Windows or OS/2 applications can use VoiceType Dictation; the dictated text can be dropped into them easily and fairly speedily even though dictation cannot be direct into the application. A PCMCIA-compliant card will be available in December and will make VoiceType portable. The card will be independent of the model used, as it is hardware-specific only. VoiceType Dictation for OS/2 is available in US-English, UK-English, French, German, Italian and Spanish. VoiceType Dictation for Windows will be available in February 1995. Both Windows and OS/2 versions run on IBM personal and personal computer-compatible desktop and mobile systems, with at least a 33MHz 80486DX or 25MHz 80486SX microprocessor, respectively. The software for both requires 8Mb memory and 32Mb of free disk space after the training is completed. VoiceType Dictation for OS/2 runs under OS/2.1 and OS/2 Warp. VoiceType Dictation for Windows runs under Windows 3.1, Windows for Workgroups and MS-DOS 5.2 or higher. MS-DOS 6.1 is required when using VoiceType Dictation with the portable card. VoiceType Dictation consists of software and a hardware board that comes with a m
icrophone. Prices for the OS/2 and Windows version cost from UKP755 to UKP855, which includes the PCMCIA card. The software on its own costs UKP381. The hardware costs from UKP465 to UKP545, which is for the PCMCIA adaptor card, which will be available in December. The radiology language model will be available in December and will cost UKP500. The other speciality ones will be out in early 1995.
