The Speech Processing Division of the Philips Company recently developed FreeSpeech 98, a new system that recognizes natural speech for conversion into text. The software package can be applied to generate and format text, while simultaneously controlling the Windows interface without using the keyboard. Philips has implemented the so-called "software only" method, which means that the programme performs the entire recognition procedure. It is now possible to have the calculations executed by a fast but simple PC processor. Currently, speech technology software has become very user-friendly. Several providers, such as IBM, Lernout & Hauspie, and Nortel are introducing their innovative products on the market.
At present, speech recognition software is supplied in the form of a separate programme for the PC. Some market experts anticipate that in a period of three years time, the application will be fully part of the operating system. The idea sounds attractive but it still remains to be seen whether this kind of integration will be authorized. One only needs to refer to the huge problems Microsoft is currently experiencing because of the integration of an Internet browser into the Windows environment. An operating system with an incorporated speech recognition package would constitute a serious form of monopoly, according to the American legislation, since the software is of equal interest to people who do not make use of the Internet.
On the other hand, if the speech recognition software is implemented in the operating system, the user can start programmes, select menus and even input data by simple voice command. State-of-the-art products are able to convert speech into text without the user having to talk in a dictating voice. FreeSpeech 98 is a self teaching programme that needs to get used to your voice. The software allows the introduction of a standard user with a male or female voice and automatically designs a voice profile for the both of them. The programme is able to distinguish between 45 phonetic groups. Initially, FreeSpeech 98 starts from a series of general voice profiles, that are obtained from of a large number of people who have pronounced the same vocabulary.
During a training session of at least 15 minutes of speech input, the general profile is adapted to the typical pronunciation pattern of the individual user. Each language requires its own recognition module in which the phonemes are identified with well defined letter combinations. English and Dutch are relatively easy to implement; German and French are far more difficult to integrate into a speech recognition system. For the Japanese language, the developers have to rely on entire syllables or symbolic signs. Today, speech recognition software is already available for specific professional use in, for instance, lawyer's practices and physician's offices. Read more on speech recognition systems for health care professionals in the VMW article Speech recognition medical specialists work more efficiently.