Just that you are standing in the concourse of Rotterdam Central Station, and you can speak into a machine to ask it the time of the next train to Amsterdam, and an electronic voice will instantly tell you the answer, including the platform number. The TU Delft Mediamatics department has been collaboring for some years with OVR (Openbaar Vervoer Reisinformatie), a company that provides public transport information, to create systems for automatic speech recognition. So far, results have been nothing to write home about, certainly not when the information was requested from noisy places like station platforms. If the voice of the passenger on the platform is drowned in ambient noise, with its mixture of announcements including delayed trains, the computer gets confused. It is an established fact that other people are much easier to understand if you can see as well as hear them talk. It is not just the deaf who use lip-reading, for people with normal hearing will also resort to watching the speaker's mouth as the level of ambient noise increases. This has led to the idea of supporting automated speech recognition systems with software for automatic lip-reading. The system could also come in useful for hands-free phone calls in cars. A small camera could be pointed at the mouth of the speaker and a processor could analyse the video images in real time. Polish IT engineer Jacek Wojdel has developed a working prototype. Automatic speech recognition has been the focus of worldwide interest for over two decades. International companies have large research departments working on it. At Philips in Aachen, Germany alone some 150 researchers are active in the field. IBM has developed the Via Voice Speech System, and the Belgium company of Lernout & Hauspie, which recently went bankrupt, was also a major player.
展开▼
