Speech technology firm tunes software

Forthcoming product releases from language and speech technology specialist Lernout & Hauspie (L&H) mark a significant improvement in technologies that have a reputation for failing to deliver.

In an exclusive preview, vnunet.com has seen demonstrations of the next release of Voice Express, L&H's speech recognition software, and Translator Pro. In both cases, L&H has tried to improve accuracy in interpreting input.

Hank Pokigo, product manager of L&H's PC applications group, said: "We are aware that expectations have been set at unrealistic levels." However, concentrating on filtering out pause sounds has produced a reduction in errors of between 25 and 30 per cent for Voice Express, he said.

In the next version, due to be released in the UK in September, L&H takes advantage of speed improvements that can be achieved using the Pentium III processor.

Apart from a much faster response to normal speech speed, the software will only require the user to 'train' the product for about five minutes.

"We believe that from opening the product box to first productive use will take no longer than 12 minutes," said Pokigo.

Other features include the ability to create a set of personal and commonly used commands such as 'open my documents', where 'my' refers to a specific Windows folder location.

Voice Express has also been tightly integrated into Microsoft Office applications, recognising common functions within the suite.

L&H said that its Translator Pro software is also more accurate. In the next version, the French, German and Spanish translation engines are said to achieve about 95 per cent accuracy in sentence translation. The software also borrows from L&H's high-end iTranslatorPro's Web translation engine that can convert text in websites.

Translator Pro is due to be released in the UK in May. Jeff Hopkins, vice president of engineering at L&H's US laboratories, said: "We are now providing more than gist translation. For relatively simple websites, this will be enough to give visitors a sense of what is being presented."

Hopkins accepted that for full commercial use customers will need to consider using additional services such as post-edit human translation. The next version will ship with a lexicon that includes IT specific words and phrases.

Speech and language technologies are making rapid advances in a market that researcher Gartner predicts will be worth more than $2bn by 2002.

Much of the growth is expected to come from the need to offer speech as an alternative input method and from the recognition that non-English speaking countries should have the means to understand internet content in their native tongues.