Nearly every one of the world's largest technology companies is trying to figure out how to let computers understand human speech, but a Santa Clara-based startup may have just cut its way to the top of the field. Hound, as its app is called, is a voice-powered digital assistant. You can talk to it, ask it questions, and have it perform tasks for you. What sets Hound apart is that it's faster and more capable than anything you've ever tried before. It's available now on iOS and Android.
SoundHound, the company behind Hound, launched the app in beta for Android users last summer, and it's spent the last eight months improving the service with the help of about 150,000 testers. The company is also launching partnerships with Yelp and Uber today to let Hound users get restaurant information and hail a ride from within the app. Those integrations are nice, but Hound has a tall order: it's trying to usurp the likes of Apple and Google as the go-to voice interface for smartphones.
>Talking to our technology is considered one of the next big leaps in computing. When software has the ability to understand what we're saying and how we're saying it, it'll be able to parse questions and supply answers, perform tasks on our behalf, and transform how we interact with devices. So far that vision hasn't quite arrived. Apple's Siri often stumbles on simple requests, while Google Now is a personality-devoid arm of the company's search engine. Microsoft's Cortana is trying to be both clever and useful, but it's virtually nonexistent on mobile phones where we need it most. Amazon's Alexa is gaining steam in the smart home, but you can't ask it anything complex.
HOUND IS THE FIRST DIGITAL ASSISTANT THAT FEELS LIKE A REAL STEP FORWARD
Hound is the first digital assistant that feels like a real step toward the future, albeit a handicapped one at the moment. It's not that Hound feels more like you're talking to a human — it's quite robotic in fact — but it is without a doubt the smartest and fastest voice-based assistant I've ever seen. The app is so fast that it can produce near real-time translations of whole sentences in other languages, and it can spit back mounds of requested data faster than you could ever possibly glean it from Google with a keyboard. You have to open the app to ask it questions, which is a drag. Although the company has added 3D Touch support to its iOS version so you can jump right into a query and a "Ok Hound" voice command to make hands-free requests.
>The software's true appeal is understanding questions within questions and sussing out human context. You can give it sprawling, absurd requests nested inside other requests like, "What is the population and capitals of Japan and China, their area in square miles, and the population of India, and the area codes of France, Germany, and Spain?" and Hound will give you the information just seconds later.
>It remembers too, allowing you to try follow-up questions. Ask Hound to find you a coffee shop within walking distance that has free Wi-Fi, and you can then tell it to exclude Starbucks to narrow the search. You can ask it for hotels costing between $200 and $400 a night in Seattle near the Space Needle and when the sun will rise two days before Christmas four years from now. All of it feels hyper-specific, and Hound's default screen is a help section guiding users on what the service is even capable of. But its underlying power is undeniably impressive.
>__SOUNDHOUND IS KEEPING ITS SECRET SAUCE UNDER WRAPS__
>SoundHound CEO Keyvan Mohajer won't disclose exactly how the company's software is able to do this when Apple and Google cannot. He says the underlying technology behind Hound is built around a unique approach to natural language processing. When combined with advances in machine learning and other artificial technology techniques, Hound is able to do what Mohajer calls "speech-to-meaning." While other digital assistant software translates what you speak into text and tries to figure out what you said, Hound supposedly skips that step and deciphers your speech as it hears it.
>"We’ve been working on it for nine years. It’s not a new direction," Mohajer tells me. "It was an original ambition of the company and we knew it was going to take that long." SoundHound has been spent the better part of the last decade as a lesser-known Shazam. The company's main mobile app identifies music and differentiated itself early on by letting you hum a tune into your phone to hear the song and artist name. SoundHound's proficiency at audio recognition has helped it license out its technology to businesses. Yet all the while Mohajer says the startup has been preparing for the moment when a digital assistant could make use of modern software advances and surpass well-known competitors.
>__HOUND DEFAULTS TO MICROSOFT BING WHEN IT DOESN'T UNDERSTAND YOU__
>It's not perfect. When Hound does stumble, it does so in weird, inconsistent ways. SoundHound built its own so-called knowledge graph — it doesn't use Google — from which the app pulls information. Sometimes that information simply isn't there. It can tell me when President Obama's grandmother was born, but can't tell me who won the Oscar for best supporting actor in 2003. It also defaults to Microsoft's Bing for anything it doesn't understand. More often than not if it's kicking you to search it's because it either misunderstood what you were asking or it just didn't know how to answer. The goal will be for Hound to avoid doing that so often that you ditch using the app altogether.
>Some of those inconsistencies will undoubtedly be smoothed over with enough time. But there will always be things Hound can't do because it doesn't have direct access to your phone's software, like Apple, or intimate knowledge of your search history and email, like Google. So right now, Hound is limited to what it's best at. For the most part, that includes finding local businesses, asking oddball queries on the fly, and doing language translation. It can also tell you how much an Uber ride will cost without having to input your pickup location and destination in the Uber app, which is a neat benefit I can see myself using all the time.
>Mohajer says the company has created an easy way to develop new "domains," which is industry jargon for a specific skill set aided by a third-party API. For instance, there's a weather domain, a hotels domain provided by Expedia, and a phone domain that each let Hound take your question or request and execute it. SoundHound says it started out in beta with around 50 domains and has grown to more than 100. Mohajer says most of the company's competitors don't offer much more than two dozen simple domains.
>Still, it's hard not to think SoundHound's technology could be so much more useful and widespread if it were adopted by Apple or Google, which would require an acquisition that may never happen. For now, SoundHound is intent on making its product the best alternative out there, and it hopes other companies will rally behind the product if it remains one step ahead of Siri and others. "I use Google Maps on iOS instead of Apple Maps, even though Apple Maps is more integrated," Mohajer told The Verge when Hound first launched in beta. "I think if you deliver something that is substantially better, people will use it." Hound isn't quite there yet, but it is most certainly on its way.
