Speech recognition firm demonstrates translator in UK

Speech recognition firm demonstrates translator in UKRichard Ball Another Star Trek fantasy, the Universal Translator, is well on the road to becoming fact, not fiction. At a speech and language symposium at Sussex University last week, Lernout & Hauspie demonstrated a prototype English – German speech translator.
“In three to four years you will be able to say something in one language, and in your own voice, the computer will say it in, for example, Japanese,” said Jo Lernout, co-founder of L&H, the Belgian speech recognition experts.
The base technology for this is already in place, and was shown off last week. L&H has speech recognition, automatic translation of the text into another language, and text to speech synthesis.
The demonstration had two operators connected via a network (simulating the Internet). One spoke in English, the other in German. The system faultlessly translated what was said by one user to the language of the other.
OK, so it’s not totally Star Trek level. For the foreseeable future it will not be simultaneous. In order to figure out the context of the words, the PC is always a sentence behind the speaker.
And it takes two or three seconds on a PC to do the translation – but it works – and that’s the important thing.
Indeed, the speech synthesis is so good that it is impossible to distinguish between a human or computer talking. If a user spends 15 minutes reading from a specific script, the PC gains enough information to allow it to synthesise any language typed into it.
“We can make you say something you never said before,” boasted Lernout. Very impressive, but slightly disturbing all the same.
“Speech technology is one of the keys to unlocking the enormous potential of computers,” said Lernout. “It will enable the number of PCs to go from a few hundred million to billions.”
But there are problems, such as the interface to computers, whether a PC or an information kiosk at an airport. “And there is too much information out there,” said Lernout. “Everyday 53,000 new users are added to the Internet. We need linguistic robots to sort through this information and mine it.”
Lernout has a vision of intelligent software that takes in a request from a human user, figures out what the user really wants and then fetches that data from the Internet.
Other technologies are being added around the speech recognition and synthesis. “We can add cameras to the computer, so it knows when you’re talking to it or to another person,” Lernout said.
Biometric verification of the face can also be used to identify the user.
“And we will add genetic algorithms to improve understanding of dialect and accents,” he said.
L&H already has the phoneme data for 12 languages in its database, and is expanding all the time.

Leave a Reply

Your email address will not be published. Required fields are marked *