The bus route which talks to you
Bente HaarstadPhoto: Jens Søraa
It is a long while since computers learned how to speak. Now they are about to learn how to listen and understand. Norway's first speech recognition machine will be used for the Trondheim bus service. But you cannot use a broad dialect when you put a question to a machine, and use phrases such as 'Wassa way to La'mon?' (Lademoen is one of Trondheim's suburbs.)
Student Jørgen Bosoni is a 'wizard' he is teaching a computer to recognize speech.
''It is quite easy to make a computer talk about specific things, but it is hard to enable it to recognize speech'', says reseacher Tore Amble at NTNUs Department of Computer and Information Science.
He is working on an interdisciplinary programme for speech recognition, and admits that his project will meet many challenges when set to work in a city like Trondheim. This is because most people express themselves by means of that peculiar set of sounds which is in Norway is known as the 'trøndersk' dialect.
Sober, refined Trondheim citizens
''Oh yes, 'trøndersk' is a real challenge. We joke that we hope all our users will be sober and refined citizens'', says Amble.The users to whom he is referring, are those travellers in Trondheim who frequently use the bus and who want to know the fastest way from one place to another in the city. Soon you will be able to ring a computer which will quickly find the best solution to your problem, and then tell you what it is. So far, however, this marvel is still at the experimental stage.
''Thank you for calling Trondheim Traffic Company's automatic route information service'', says a staccato voice just after having rung the system's test number. So far there is no computer which can provide information about routes; instead there are students who pretend to be a computer. And actually, this functions well. It is almost as if you can hear the computer brain thinking once you have asked for the fastest bus to whichever place you want to go, but it certainly must feel odd to pretend that you are a machine in this wonderful computer age.
Training the computer
''These simulated tests are very important, and they involve 'training' the machine. Take for example the word 'Risvollan' (local name on a suburb). The machine will not recognize this word until we have taught it that a particular sound means just that'', says Tore Amble.At this point the automatic route information consists of a telephone number at NTNU. Students at the Department of Linguistics are positioned here, and they are called «wizards» by those who participate in the project. Anything this hypothetical bus traveller can ask is answered by these wizards, recorded, and then fed into the computer's training programme.
It is so long since computers became such competent talkers that we hardly realize that what we hear is an artificial voice, when we ring for instance, the bank account phone-service. Though the voice itself is not artificial, it has been put together artificially and what we hear is bits of speech which have been spliced together.
This means that to teach machines to understand questions is much harder than to teach them to talk. But in a few years, possibly just into the next millennium, the computer will be able to handle all the problems which a bus passenger might have in Trondheim.
A lot of support
The first speech-recognition programme started last year, and it is planned to last for three years. In addition to the Department of Computer Systems and Telematics, the Departments of Linguistics and also Telecommunication are involved. Torbjørn Svendsen from Telecommunication is in charge of the project.Research into speech-recognition machines has been found to be of such interest that the Norwegian Research Council is supporting yet another programme within the same field, to the extent of NOK 6 million. Such interdisciplinary projects are prioritized within the new university strategy at NTNU.
Though some may question the need for a technology which will obviously make several employees redundant, the leaders of the project refer to the Oslo area traffic-information service, which receives more than 25,000 calls a week. In Trondheim the pressure on the information service is not as great as that, but the project has been placed here to make it as good as possible in its initial phase.
''At NTNU we need to view technology with a critical eye. We should certainly always ask ourselves three fundamental questions such as: is a new technology possible, is it superfluous, or do we need it? Loss of jobs is probably crucial when dealing with computer processing and automation. However, the aim is not to make people redundant, and it should be a positive thing if people do not have to do things that are merely stressful and are not very interesting. One can draw a parallel with the automation of telephone exchanges'', says Tore Amble.
Try it yourself
If you want to try the simulated automatic route information service, the telephone number is + 47 73 59 77 25. This service is available on Tuesdays from 12.00 noon to 2 p.m.
If you use the Internet, you can try written questions at this website: