12.1. Introduction

The past decades have witnessed impressive advances in the development of automatic speech processing. The VIP100 (the first recognition system to be commercialized in 1972 by Threshold Technology Inc.) could only recognize 100 isolated words after a long and tedious training phase and occupied a volume of approximately one cubic meter. Nowadays, adaptive systems may be integrated into a portable computer and are able to recognize virtually unlimited vocabularies.

At the end of the 1970s, when microchips appeared on the market, the first results obtained in automatic speech processing were so promising that they triggered very ambitious predictions: a journalist claimed that typewriters with keyboards would totally disappear by 1986 and that they would be entirely replaced by vocal interfaces. We are still far from achieving this today, but a number of evolutions have taken place, one of the milestones being the presentation by Philips of the first (almost) real-time continuous speech recognition system in German, at the Eurospeech conference in Berlin in 1993 [EUR 93]. Since then, sharp competition has been observed between research groups and companies, leading to a significant drop in product prices and a noticeable improvement in their performance and usability. Voice technologies are now not only used by professionals, but also by the general public: voice servers are one example.

With the advent of the Internet, anyone from anywhere in the world may access any remote database, whilst communicating with several other users, and...