Phonetic Systems: Like talking to the operator again

Last week, Petach Tikva-based Phonetic Systems announced the completion of a $37.5 million financing round. The round was in fact in two stages, the second of which yielded “only” $27 million, but it included some well-respected investors. The round was led by US fund General Electric Partners, considered the world’s largest fund for Internet and IT investments.

Phonetic refuses to disclose exactly how much the fund invested, but from the nods in response to suggested figures, it is presumably more than $15 million. The investment is reported to be the fund’s first in Israel, and joins an investment in the company by another giant, Sonera of Finland.

”I believe it again proves the added advantage of receiving investments from funds. Throughout our activity, we’ve had offers from various private investors and angels based on high company values. By choosing funds we were able to successfully reach giants like General Partners and Sonera,” executive VP business affairs Hezi Resnekov says.

Resnekov was VP R&D Atzmon Gilai’s first boss in Amdocs (NYSE: DOX) and in the early 1990s, they first encountered large database directories, like the Bezeq information service.

”It’s a service with serious peak problems. At certain hours there’s a shortage of people, while at others, there are too many,” Resnekov says.

Peaks and troughs

50,000 people are employed in this field in the US, leading to severe financial and managerial problems. “Suppose you need to change shifts. It’s impossible to suddenly change 100 people for another 100. It needs to be done gradually so that the level of service is not affected. The market appears to be very suited to automation. At one time, a great deal of money was invested in the area, and the feeling was that, even if there were losses, the problem could be solved. Our initial assumption was that it was impossible to solve the problem through conventional voice recognition technologies. When we started, the best technologies were capable of recognizing only 50 words, and we’re talking about technologies expected to support hundreds of thousands. Our view was that it should be approached differently.”

Separating ear from brain

Phonetic decided to differentiate between listening and understanding, in other words, separating the ear from the brain. Resnekov: “When you first hear a sound, you don’t immediately discern what is said. You know only understand after the sound has undergone processing and been compared with the database in your brain. We decided to concentrate on voice recognition relating only to the sounds of the language, not content, to take them through a process and bring them to the brain to check if it recognizes them. This is why we call ourselves a voice search engine, not voice recognition.”

In Phonetic's system, the technology does not relate to the size of the database, since it does not deal with the meaning of what is said. “Only 5,000 words are necessary in order to speak good English. Webster’s dictionary has 90,000 words. But there are 1.5 million family names in the US. That should give an idea of the size of database we are talking about.”

From smart computer to telephone

It took Resnekov and Gilai almost two years to prove their initial theory. They left Amdocs in 1994 and started working together. Gilai: “At the time, speech recognition technologies were very primitive and mostly focused on limited vocabulary, aimed at giving commands to various appliances, such as forward, backward, next, previous, etc."

Resnekov: “I think the situation was even worse then, since several attempts had been made at the end of the 1980s which failed miserably. Only AT&T had some measure of success with a system that recognized about 15 words. US venture capital funds invested huge sums in these initiatives, which killed the market for ten years. Some people say they are still licking their wounds.”

Resnekov and Gilai raised their first $1.4 million from Evergreen and what is now called Sadot R&D. Resnekov: “There was an obvious market and a technology that needed to be proved. The trend among venture capital funds was to start investing only after you start selling, and they became known to us as “non-venture capital” funds.

The company decided to enter the market via large enterprises (3,000-5,5000 employees), rather than communications companies in order to find early adopters and to avoid touching the open wounds left by the earlier unsuccessful attempts. Phonetic claims many Fortune 500 customers, and again declines to disclose their names, although it says it has 60, some of which have had their product installed for more than two years now. The largest system serves 200,000 postings and a 2 million posting capacity is currently being tested. Resnekov says installation is not time-consuming, taking an average one-and-a-half days for a 200,000 strong organization. “It’s almost an off the shelf product,” Resnekov says.

Gilai: “Systems like ours have to be user friendly. It doesn’t matter how efficient they are, people will not use them if they are not user friendly.”

Competitors, but not serious

Concerning competition, Resnekov says, “Two listed companies SpeechWorks and Nuance, currently worth $1.5 billion each (worth much more a few months ago), reopened the market, but they haven’t solved the problem of the database. There are also proper voice recognition companies: IBM’s Dragon and ViaVoice (which was acquired by failed L&H – G.N.) with Dragon Dictate, but these are products for the individual user that require quality microphones and software training for it to recognize your voice. This is not our field.

”There are other solutions in Israel as well, in cellular telephony,” Resnekov adds. “Cellcom has a system on its server, which compares every new instance with the original recording. Some of Pele-Phone and Partner’s handsets have them. I see Phonetic as one of four global leaders in this field, together with SpeechWorks, Nuances and Philips. Other small concerns exist, none of which is developing new technology. Don’t forget that this technology has been in the works for 40 years, a great deal of money has been poured into it and the smart thing to do in deploying systems like these is by way of evolution, not revolution – not by trying to solve the problems of the human brain, but by gradually providing solutions and improving them with time.”

Phonetic’s world is based on large organizations. Resnekov: “We were surprised to find this market interesting not only as a springboard for proving our technology. These organizations have employee databases that constitute a basis for further developments and once they are adapted for speech recognition, other things can be added, such as meeting schedules, voice use of Outlook, Intranet, and so on.”

Another potential market is carriers. Resnekov and Gilai believe that voice dialing will follow voice search engines. “See how the wheel turns. I used to ask the operator to connect me to someone, later I used telephone numbers as we developed ringing tones and we’re now developing backwards to an era in which we will once again talk with the system.”

Phonetic currently has an agreement with Avaya (struck when it was still a Lucent division), which serves as a US marketing arm. It is in advanced negotiations with another international concern marketing to enterprises. Gilai assesses that its products are installed in 75% of the market for such products and believes that all Fortune 1000 companies will start using voice-based systems. Disappointment awaits anyone waiting for a Hebrew version of the solution. “There are several reasons for this: the market is too small; Israel does not have an accessible database of recordings for initial work to be carried out; and there’s the problem of pronunciation of Hebrew vowels that have no equivalent in English,” Resnekov says.

As for going public, Gilai and Resnekov prefer to continue with sales, and perhaps consider it next year. “The funds are not pressuring us, so why should we feel pressured?”

Published by Israel's Business Arena on 27 February, 2001