The speed and accuracy of Soundhound’s in-car voice recognition floored me.
A new voice recognition technology that could be in your next car should have Apple and Google worried. It's called Houndify, and it lets drivers pull up information like driving directions and weather simply by asking.
This in and of itself isn't exactly groundbreaking; you can of course pull up the weather by asking, say, Siri or Google Now. But it's the speed and accuracy of parent company SoundHound's technology that completely floored me when I met CEO Keyvan Mohajer for a demo.
"We can do speech recognition and natural language understanding simultaneously," Mohajer told me on the CES show floor on Friday morning. "We call it 'speech-to-meaning.' What Apple, Google, and [voice recognition pioneer] Nuance do is speech-to-text and then text-to-meaning."
This technology has been in development for 10 years, Mohajer said, adding that it's unlikely that too many other startups will attempt to break into the space because it takes several years to develop the technology. (The company was founded by a couple of Stanford PhDs.)
Not only does breaking a user's voice request into two operations make the whole process slower, Mohajer said, it also makes it less accurate because "if you're doing speech-to-text without caring about the meaning it's more likely to make a mistake," he said. "But if you care about the meaning from the beginning you can reduce errors by quite a bit."
Sounds great, I said, now how about a demo?
Mohajer was more than happy to oblige, as seen above. In the demo, Mohajer asked a mock car dashboard a series of increasingly complex commands like "What time is it?," "What time is it in Tokyo when it's 2pm in San Francisco, California?," and "Show me hotels in San Francisco for tomorrow staying for two nights that cost between $200 and $400 per night and are pet friendly."
All of these commands were answered by Houndify nearly instantly, without the pauses you might find while Siri is "thinking" about an answer.
Nvidia, the company best known for its line of high-end graphics processors, will integrate the Houndify technology into its Drive CX dashboard platform, which several car manufacturers, including Audi, BMW, and Mercedes, have already committed to supporting. And while the Houndify technology works best when there's an active internet connection, it will still be able to perform certain in-car functions, like opening doors and windows, while offline.
While SoundHound was at CES to specifically discuss its in-car technology, the underlying technology is also being integrated into Android and iOS mobile apps called Hound that are currently being beta-tested. These apps should be released soon, I'm told, giving more people the opportunity to try out the impressive voice recognition tech.