The W Project

Towards a Free Multilingual Talking Computer for the Disabled

GOAL

The W project aims at creating a fast computer keyboard driver for people with speech disabilities. The related software is based on grade II Braille languages developed by blind people associations all over the world and minimizes the number of keystrokes to utter a word (the name of the project is the grade II abreviation for "word" in English). In combination with the MBROLAproject, W could put words in the mouth of thousands of users, in tens of languages.

People working in the area of speech synthesis are frequently asked how and when their research will be put to use for helping people with visual or speech production disabilities. While significant efforts have been made these last years towards shortening delays between research and production in this area, one must admit that people with speech disabilities are frequently left aside. One of the reasons for this is undoubtedly that real-time interfaces to TTS systems remain a problem: even a trained secretary cannot type as fast as one speaks.

Existing solutions, namely sentence assembly (using a specially designed hardware that replace traditional keyboards, or software tools to emulate such hardware), word prediction, and abridged languages, have pro's and con's.

As a result, we have chosen to explore abridged languages and more particularly Grade II Braille for that purpose. Grade II Braille has been in use since 1829 to help blind people read and write texts in a concise way. It differs markedly from Grade I Braille, in which each character is mapped to 6 dots (no contraction is used). For English, for instance, Grade II Braille provides 189 contractions, including contractions for sequences of characters that are frequently encountered. It provides contraction ratios of 30 to 50 percent, depending the type of text involved. This result is similar to that provided by word prediction algorithms, but contractions are much easier to use than word prediction systems, for which users have to check predicted words; contractions are systematic enough for human users to learn them by heart, which has been done extensively by blind people for the past hundred and seventy years).

Another problem when trying to use a TTS system on a real-time basis is that intonation and rhythm generation has to be adapted (words have to be spoken as quickly as possible, while maintaining as much quality as possible). We therefore study this issue.

MEANS

The W project was initiated in June 98, with a first implementation of these ideas for French : WFRENCH(in French)(in English). The reliance of the resulting program on widely used engines -- Perl for contraction rules, trainable CARTs for intonation and rhythm, and MBROLA for synthesis -- makes it easy to port to other languages, thus providing both a working application and a portable kernel for further testing and development of the use of abridge languages to facilitate the use of speech technology by handicapped persons.

(This project benefits from the kind collaboration of O. Platteau, our official beta-tester)

ONGOING DEVELOPMENTS

The W software has recently been turned into HOOK, a Windows keyboard hook, which captures keystokes from any applications, and does the abrevaition expansion.

We are now also part of an EC IST5 Project for word prediction and abreviation expansion : FASTY.

Last but not least, we are working on adapting W and HOOK to the Elena stenographic language.

Last updated December 17, 1999, send comments to dutoit@tcts.fpms.ac.be