Talk back

Text-to-speech software is becoming more mainstream. Which European companies are at the forefront of the technology?

It is an age-old problem. The most effective way to build deeper relationships with customers is to talk to them - but the profit margins on many company's products and services will not support human agents.

The obvious solution is automated voice systems, and an increasing number of companies are starting to use a combination of speech recognition, speech verification and speech synthesis technologies to interact with their customers (see Infoconomist, February 2001).

To date, the least exploited of these is speech synthesis. This takes the results of a database query, often activated by a customer asking for something over the phone, parses the text, analyses its syntax and intonation against stored models, and then delivers back a response in a human voice. "Until now, the quality [of speech synthesis] has been lousy," says David Bradshaw of market analyst company Ovum, but he says, recent improvements in the underlying technology have made its application more compelling.

Nuance and Speechworks, the two leading speech recognition vendors in the US, have both been upscaling their speech synthesis offerings in recent months. Nuance launched Nuance Vocalizer, a suite of text-to-speech (TTS) software products, in January 2001. SpeechWorks, meanwhile, acquired speech synthesis specialist Eloquent Technology in December 2000, and is in the process of adding 12 additional languages to Speechify, its TTS software offering.

There are, however, a number of European contenders down at the base technology level. Given the complexity of speech synthesis most of those players emanate out of university research laboratories. They also have little - if any - venture capital funding, and often rely on partnerships with other technology vendors to flesh out their offerings.

Babel Technologies, a spin-off from Belgium's Faculté Polytechnique de Mons, is probably the most well known of Europe's smaller players. Founded in 1997, the company's core product is Babil SDK, a set of text-to-speech (TTS) software development tools, based on the MultiBand Resynthesis Overlap Add (MBROLA) speech synthesis engine developed at Mons.

MBROLA's 'diphone concatenation' software algorithms - the sound generating element of the software engine - enable Babel to supply its customers with more life like speech than its competitors, says Vincent Fontaine, CEO of Babel, which is particular important for customer relationship management style applications. MBROLA is available in nine different languages, but the company does not build applications itself. Its core customers are system integrators, although it also targets large telecommunications companies and voice portals.

Another major centre of speech technology development in Europe is the Swiss Institute of Technology. In 1998, work at the university led to the creation of Svox, a Zurich-based spin-out targeting third party application developers. The company's core technology is called SPLISS, on which its suite of embedded TTS software, eponymously called Svox, is built.

Volker Jantzen, CEO of Svox, says that, like Babel, his company will target its products mainly at system integrators. But, he says, "I expect the biggest demand for Svox's technology to come from telcos and big [voice] portal operators. Building email reading systems will be a key driver behind telcos licensing components of our TTS software".

Rhetorical Systems, a spin-out from Edinburgh University, is taking a different approach. It plans to offer end users a full suite of TTS software applications based on its rVoice speech synthesis engine. To help companies develop tailored speech systems, it also offers consultancy services and will also host software on behalf of customers on an application service provider basis - a model that is proving popular in the US.

The company, which was founded in April 2000, has received EU4.8 million in first round funding from private sources, and plans to roll out the first products in mid 2001. According to Marc Moens, CEO of Rhetorical Systems, the first iteration of the software will support English language only, although there are plans to add support for German, French, Dutch and Spanish by the end of 2001.

The other key contender in Europe's speech synthesis sector is Elan Informatique, which was acquired by troubled speech giant Lernout & Hauspie in March 2000. Elan's core product is Speech Cube, a suite of well regarded server-based multilingual TTS software. The technology is used by a number of telecoms companies and web site operators.

"Last year's [2000] killer application was reading email, such as delivered by Alcatel [the French telco], but now demand for personalised web content is getting stronger," says Etienne Lamort de Gail, director of marketing, at Elan Informatique. However, given the financial difficulties of L&H, there is speculation over whether or not Elan will remain part of L&H, or whether it will be sold off or spun-off as a separate entity.

Author: James Thompson Date: 22 February 2001 infoconomy 2000