CeBIT 99: Speech Recognition Advances in Quality and Applicati

There is probably no sector of the multimedia market that has been more transformed in the last 13 months than speech recognition. Changes include the way voice recognition is used on the PC, a marked improvement in recognition quality, and the widespread introduction of automated systems in European telephone transactions. 1998 was the year that we turned from discrete speech to real or continuous speech. The PC market is owned by IBM with its Via Voice, Lernout and Hauspie with Voice Xpress and Dragon Systems with Naturally Speaking. All three packages are continuous speech technology. All were released this year, and all retail for under $100, including a microphone. Voice Xpress and Naturally Speaking can reach speeds approaching those of a good typist. All include natural language commands. This will be a key feature in the future development of the PC interface. Natural language interpretation, married to voice recognition linked to digital information stored with digital content could make for a formidable revolution. The technology for this is just a year or two away, but the content format has yet to be defined. L&H will be debuting an integrated dictation and language translation package. Using the companies existing speech recognition technology, the software will instantly translate the spoken text into German, French, Italian, Spanish or Portuguese. The company also says it will demonstrate a German version of what it calls its "near human like" computerised voice, called RealSpeak. The other revolution in speech recognition has been in industrial-grade telephone systems. All over Europe organisations like cinema chains are installing systems that can choose from limited lists of possibilities. This type of limited speech recognition can run on very cheap digital signal processing (DSP) chip-based cards. The dedicated processing strategy of the DSPs offers massive cost savings and massive performance improvement over a more broad general PC approach.