Archive for May, 2005

Scansoft and Nuance to merge

Tuesday, May 10th, 2005

Read the press release on Scansoft’s website.

Voice World Europe 2005 (London): speech technology more than ready for prime time

Sunday, May 8th, 2005

Of all the European speech and language technology events that I have visited over the last years, Voice World Europe 2005 has been one of the most informative. The combination of a business-focused conference with a technology-focused exhibition (annex product-focused seminars) provides a perfect blend for all parties involved in developing and deploying speech applications in a real business environment.

Let’s start with some impressions from the conference.

As a speech application professional based in the vicinity of Brussels, I was more than interested in what Belgacom had to say about Lisa (reachable at 1234) , their half-year-old automated directory assistance line. Personal tests and press reports had already indicated that Scansoft’s implementation of the system had been somewhat … suboptimal, to say the least. Philippe Bovy, Directory Assistance Manager from Belgacom, was frank enough to implicitly admit that a success rate of 50% is well below standards: their goal is now to achieve 70% by the end of this year. To Belgacom’s defence, it should be said that of all current speech enabled telephony applications, automated directory assistance is among the most difficult. Indeed, recognising hundreds of thousands of different proper names (people and addresses) is far from trivial. On top of that, the multilingual aspects of the Belgian environment make this inherently gargantuan task even more complex. Mr. Bovy concluded his presentation by urging his audience “to rely more on customer feedback than on the supplier’s advice”. Was I the only person in the conference room to interpret this diplomatically stated comment as a sneer to Scansoft?

A personal note: irrespective of Lisa’s design mistakes or child diseases, my message to potential Belgian users of speech-based telephony applications is this: don’t consider Lisa as representative. Firstly, many applications (e.g. flight information lines, phone banking, automated attendants, appointment schedulers, … ) have smaller grammars, which means that the inherent speech recognition subtask is less complex. Secondly, speech recognition accuracy is only part of the story: success also depends on the Voice User Interface (VUI).

Numerous speakers (a.o. from Postbank, Bell Canada, Mobilkom Austria and Merrill Lynch) indeed stressed Voice User Interface Design as a critical success factor. Another recurrent refrain was caller expectation & perception management. Do include a marketing representative in your speech project steering group! Belinda Banks from Bell Canada spoke with passion about Emily/Emilie, a bilingual (but virtual) customer service representative that has picked up tens of millions of calls for Bell Canada for the last two years. Ms. Banks pointed out that although on average actual call duration has increased, the customer’s perceived time on the phone has decreased. Callers apparently do associate queue time to call time – rightly so, I may add.

Noel Quaedvlieg, Head of Telebanking and Telesales of Postbank (Germany), and Terry Kaye, Head of Nationwide Direct (Nationwide Building Society, UK), explained how they overcame scepticism and bad past experiences w.r.t. speech technology in their organisations. Both companies have worked with external speech experts (resp. VoiceObjects and Syntellect/Nuance) to migrate from a DTMF-based IVR approach to state-of-the-art voice portals, thereby achieving higher automation rates. About this last aspect, mr. Kaye pointed out that “there are limits in productivity improvements by customer service representatives”.

Henrik Thomé, CEO and founder of Envox, gave examples of speech projects with ROIs of 6 months (Cyracom, US), and less than 2 months (Apoteksbolaget, Sweden). The last example shows that a “small” country population – there are about 9 million Swedes – does not preclude speech projects with fast ROIs.

All of this makes me wonder how much low hanging fruit there is on the Belgian market (10 million inhabitants) right now, just waiting to be picked.

So far for the conference. At the exhibition, I was quite impressed by Loquendo’s new “emotional” TTS voices. It seems like Scansoft, Acapela, Nuance & others have some catching up to do. On the voice application development front, I checked out the latest novelties at the booths of DBScape, VoiceObjects, Envox and Vicorp. In this area, I have some catching up to do: test the newest software. Maybe something to report on in a following post.

I was told that attendance at Voice World London 2005 was lower than previous years, both in terms of exhibitors and visitors. Anyway, I was happy to be one of them. This conference & exhibition are more truly European than e.g. the Salon Européen des Centres d’Appels (SECA) in Paris. Quite ironic, that London should be the location for a European encounter of this kind. Also, the clear focus on speech technology is just right for me.

To conclude: as far as I’m concerned, speech technology is more than ready for prime time, not just in the large English-speaking markets, but also across Europe, including my own country.