Two vendors of mobile voice technology have teamed up to power carrier services that transcribe voicemail messages and conference calls into text. Ditech Networks, which supplies gear that scrubs background noise from cell-phone calls, signed an exclusive deal with SimulScribe to resell that company’s voice-to-text technology and integrate it into fledgling mobile Web-based services.
Voice-to-text services pose obvious advantages for users who want to send a written message when they don’t have access to a keyboard and to recipients who lack the time or inclination to listen to their messages. But accurately transcribing spoken words has been a serious computing challenge. Ditech said it believes SimulScribe has the best available translation technology, but even with it, the companies will be offering an operator-assisted system in addition to full automation.
As a voice-processing equipment vendor, Ditech has relationships with mobile operators including Verizon Wireless and AT&T.
More from Network World Canada Wireless operators promise ‘simple language’ in contracts
A Ditech spokesman said the company also has relationships with Canadian carriers and plans to make the service available in Canada. It now wants to sell carriers the tools they need to offer their subscribers a voicemail transcription service that costs less than using call-center operators. As an add-on service, having operators listen to voicemails and type them out typically costs between US$5 and $25 per month, depending on the number of messages, according to Karl Brown, vice president of marketing at Ditech. A fully automated service would probably start out costing about $5 per month and fall in price, eventually reaching the point where carriers can bundle it for free, he said. Ditech will also make the technology available to enterprises to transcribe their own voicemails internally.
Ditech will integrate SimulScribe’s technology with its own mStage voice-quality product platform, which will improve the accuracy of voice-to-text in typical usage environments, the company said. The company also plans to integrate SimulScribe’s technology into a suite of tools, called Toktok, that it offers today in a beta test. Toktok, a set of mobile Web-based applications, includes capabilities such as voice-based call commands, “whispered” audio notifications of calendar entries and social-networking messages, and memo dictation over the phone. It probably will emerge from beta in the next two months, Brown said.
Ditech will immediately resell three SimulScribe services. SimulAuto is fully automated and designed to handle the kinds of words used in typical voicemail messages, offering the highest level of privacy because no operator listens to the message, Brown said. SimulHybrid can send parts of a message that the software isn’t sure about to a human operator to ensure greater accuracy. ScribeAll is designed for real-time voice-to-text conversion of longer messages or conference calls, with or without an operator. SimulScribe’s technology supports multiple dialects of English as well as Spanish and is available around the world, according to Ditech.
“People have been trying to do voice-to-text forever,” or at least for the past 20 years, said mobile analyst Jack Gold, of J. Gold Associates. Accents are an issue, as well as background noise. “It works OK if you’re going to do command and control. It’s very hard to do if you’re trying to do complete dictation,” Gold said.
With enough computing power, this type of application is possible, Gold said. But even with 99 percent accuracy, one word in 100 could be wrong, which would reduce its usefulness, he said. “You’ve got to get it close to what you and I would be able to do in a conversation, understanding-wise,” Gold said.
For the exclusive rights to sell the voice-to-text technology, Ditech will pay SimulScribe $3.5 million upon signing and another $3.5 million at the end of the second year of the relationship. SimulScribe can earn an additional $10 million over three years, the companies said.
With files from Greg Meckbach
