Syntiant – Always-On Voice AI Chips at The Edge
Table of contents
The ability for marketers to gauge intent these days is spooky. Performing a simple Google search for “hotels in Angeles City” while sitting in a cafe in Manila will suddenly surface “cheapest transport from Manila to Angeles City” ads in your Facebook stream. It knows you’ll need cheap transport to get there so you can spend your money on other things. What you may find even more surprising is when you’re talking to a mate on the phone about the carnal pleasures of Angeles City and suddenly STD test ads start appearing in your Twitter feed. Is your phone really listening to what you’re saying?
Turns out that your phone is indeed listening to what you’re saying. It’s something we covered in our piece on How Audio Beacons Monitor You Via Smartphone. The truth is, there are companies that skirt the grey areas and use what you say to sell you more stuff. As voice command interfaces become more pervasive in our tech ecosystem, companies are figuring out more ways to utilize the technology. Today, we’re going to talk about a company that’s helping voice command technology go mainstream.
About Syntiant
Founded in 2017, Los Angeles startup Syntiant has taken in just over $65 million in funding from Intel, Microsoft, Motorola, and the Amazon Alexa Fund. That money is being put to use to bring deep learning to devices from earbuds to mobile phones – artificial intelligence (AI) at the edge. Syntiant’s ultra-low-power AI processing solution enables Always-On Voice (AOV) control for most any battery-powered device, from earbuds to laptops.
Just this month, Syntiant announced the availability of its Syntiant® NDP120™ Neural Decision Processor™ (NDP). The NDP120 applies neural processing to run multiple applications simultaneously while consuming very little battery power (a key requirement for devices at the edge). The list of functionality includes:
- Echo-cancellation – what it says on the tin
- Beamforming – method to enhance speech recognition
- Noise suppression – similar to the great work they’re doing over at Krisp
- Speech enhancement – tone down heavy accents or make one’s speech more clear
- Speaker identification – identify a person by voice
- Keyword spotting – identify certain words being spoken
- Multiple wake words – allows a system to be using no power and only turn on when a wake word is spoken
- Event detection – gunshots, car crashes, vehicles starting up
- Local commands recognition – being able to understand commands in any language
There is much more to voice commands than just asking Alexa trivia questions. For example, when you are listening for a wake word, you’re on all the time, and consequently, any number of false alarms can happen. When you’re looking for keywords, the device is looking for words spoken in a limited range of sounds, so false alarms are less of a problem. In either use case, you need to assure a very high degree of accuracy for usefulness while using as little battery power as possible.
Syntiant’s chip was built from the ground up so that it could do just that – perform a complex set of tasks while using minimal amounts of power. At the core of their technology is the Syntiant Core 2™, a highly flexible, ultra-low-power deep neural network inference engine, that can shorten time-to-product by months or years as compared to more constrained and power-intensive solutions.
10 Million Units of Traction
The company’s first-generation neural network, Syntiant Core 1™, has already sold more than 10 million units. That’s incredible. A startup that went from idea to shipping ten million units in just over 3 years. Even the first-generation product is impressive, offering 100x efficiency and 10x the throughput over traditional CPUs and DSPs, enabling battery-powered devices with AI functionality that doesn’t require a cloud connection.
As with most products, increase all the numbers by one and you get the next generation. The Syntiant NDP120 (which contains the Syntiant Core 2) is sampling now and will be shipping in production volumes in summer. The price? If you order 10,000 of them, they’ll cost you about $6 a piece. At that price point, soon everything around us will be able to understand what we’re saying.
The Human-Computer Bottleneck
His Excellency Elon Musk is often misunderstood by the media and the “billionaire man evil!” placard wavers. To better understand Mr. Musk, watch Episode #1169 of The Joe Rogan Experience where he says some very profound things about the interface between computer and man. Right now, your ability to communicate with the machines is a keyboard which you peck away at like a chicken. The speed of communication is atrocious. Once this bottleneck is removed, the world will quickly become unrecognizable as man melds with machine. Using voice to communicate with machines is one step closer to the brain-computer-interface where your thoughts will instantly be rendered.
Perhaps the best use cases for always-on voice haven’t even been thought up yet. All you MBAs out there ought to start thinking about what you’d be listening for if any battery-powered device could listen. Some interesting use cases can be found in the shadowy depths of the Defense Advanced Research Projects Agency (DARPA).
The Walls Have Ears
Battlefield sensors that live on dime-sized batteries can detect things for about a month before the battery runs out. That was until DARPA initiated the Near Zero Power RF and Sensor Operations (N-ZERO) program which extended the battery life to four years using advancements in technology. Having soldiers out in the battlefield replacing sensors all the time is dangerous stuff. In the world of war, it’s all about doing things out at the edge with minimal battery consumption using zero-power or low-power sensing devices.
Recently, the Chief Scientist at Syntiant, Jeremy Holleman, directed the “Near-Zero Integrated Analog Classifier” project as a result of a grant from DARPA N-ZERO. As a result of the project, they “demonstrated an integrated system capable of detecting and classifying and differentiating the presence of a generator from background noise with 100% probability of detection and zero false alarm rate while consuming 7nW of power.” If they’re able to detect when someone turns a generator on, they can probably detect all kinds of acoustic signatures, even particular vehicle types. It’s easy to see where the security applications come into play here in military and commercial settings.
An interesting side note for investors in Teradyne (TER), a stock we hold in our own Nanalyze Disruptive Tech Stock Portfolio. We originally learned about Syntiant while researching Teradyne and coming across a press release titled “Teradyne and Syntiant Collaborate to Significantly Shorten Time to Market for Innovative Artificial Intelligence Neural Decision Processors.” From the press release:
Using its UltraFLEX test platform, Teradyne is currently supporting the qualification and production ramp of the Syntiant® NDP120™ and Syntiant® NDP121™ Neural Decision Processors, Syntiant Corp’s second generation hardware platform to run multiple audio applications simultaneously at under 1mW power consumption.
Credit: Teradyne
That’s good news for Teradyne if Syntiant really starts selling lots of chips, though we have no idea what the impact on revenues would actually be. Given that Teradyne thought that was worth a press release has to mean something.
Conclusion
The ability for any battery-powered device to listen and understand what’s being spoken may sound a bit scary because it is. You can be sure that the compliance folks have their work cut out for them making sure that everyone is kosher with how their devices are used when it comes to privacy. Get past these concerns and you’ll see a world of opportunities for chips that understand the world around them by listening.
Sign up to our newsletter to get more of our great research delivered straight to your inbox!
Nanalyze Weekly includes useful insights written by our team of underpaid MBAs, research on new disruptive technology stocks flying under the radar, and summaries of our recent research. Always 100% free.