Snips AIR: a Private-by-Design, Open-Source, Decentralized Voice Assistant
This is part 1 of our token sale announcement series. To learn more, join our Telegram group!
--
Hop over to all posts about our Snips Token Sale: https://medium.com/snips-ai/token-sale/home
Radios, TVs, washing machines, computers, music players, smartphones, smart speakers… Technology is now a central part of our life. The promise is simple: make the effort to learn how a machine works and it will do something for you more efficiently than you would yourself. Yet, for all the magic and progress it created, a darker side emerged: we became addicted. We kept wanting more technology to do more things, forcing our brain to create coping mechanisms to avoid feeling overwhelmed.
But what if getting machines to do things was as simple as asking them directly? If we remove the cognitive effort of learning how they work, then there would be no friction, no sense of saturation. We could be hyper-connected and yet feel disconnected, making smart devices as ubiquitous and intuitive as electricity. Artificial Intelligence can make technology so effortless that it disappears from our consciousness.
The Snips Voice Assistant
We created the Snips assistant with one goal in mind: to build the first ethical voice assistant. Indeed, most popular voice assistants today represent everything that is wrong with the old internet: they are closed source, offer zero Privacy, centralize all the data into a server and have a business model that relies on exploiting personal data. And although they pretend to be platforms anyone can build on top of, they are actually aggregators, using the data from all the voice apps people create to capture all the ecosystem’s value.
This is ultimately the most important distinction between platforms and aggregators: platforms are powerful because they facilitate a relationship between 3rd-party suppliers and end users; aggregators, on the other hand, intermediate and control it. — Ben Thompson, The Bill Gates Line
The Snips Voice Assistant is the exact opposite. It is built with Privacy by Design, is open source and decentralized. Instead of running in the Cloud, it runs fully on the Edge, processing your voice directly inside the device you are talking to. Everything, from the wake word to the speech recognition, natural language understanding and dialog runs locally on-device. No data is ever sent outside your home.
This 100% on-device approach has a number of advantages, from enabling offline use cases, to having no network latency and being resilient to cloud outages. The fact that Snips is Private by Design therefore also means it is natively GDPR compliant. And since processing is decentralized, it is also resistant to mass surveillance and data breaches.
Snips currently understands English, French, German and Japanese. Spanish, Portuguese, Italian, Mandarin and Korean are in the works, with more languages planned for 2019.
We are also committed to making Snips open source over the next few months, starting with our NLU engine. This will improve transparency, while enabling the community to participate in the development of ambitious new features.
Read our technical paper on embedded voice inference
Snips AIR
Like any other, our voice assistant needs a device to run on. Although many people are already using Snips on their Raspberry Pi, Debian computer or Android and iOS phones, making other voice assistants disappear requires offering a delightful, simple product that anyone can use.
Today, we are announcing our first consumer product, Snips AIR. Rather than a single device, Snips AIR is a network of devices collaborating to form a mesh of AI assistants. This enables your entire home to be seamlessly voice-enabled, while protecting your family’s privacy. As Snips AIR runs locally without any voice data being sent to the cloud, no-one can spy on you or your kids, keeping your home private as it should be.
Snips AIR devices communicate with one another using a shared protocol (the AIRnet protocol), which consist of both a network abstraction layer and an ontology layer. Any device manufacturer can implement the open-source AIRnet protocol and make their device plug and play with Snips AIR. Such devices are called AIR devices.
Designing a voice assistant as a network of devices has a number of advantages over standalone smart speakers, from enabling sharing of resources such as microphones and processing power, to enabling multi-modal interactions.
Indeed, voice alone is not enough, as sometimes it is best to touch or see. For example, when asking for restaurant recommendations, it is more convenient to see them displayed on a screen rather than listed by voice. This would be easily achieved with Snips AIR, as you could ask your smart speaker to show recommendations on your TV. Since both would be connected in the same mesh, they would seamlessly interoperate.
Anyone can build an AIR device, either by implementing the underlying protocol, or simply by integrating the Snips voice assistant in their product. Coffee machines, TVs, speakers, lights, they could all be part of the AIR network and share their resources. At Snips, we are working on two AIR device concepts: a smart speaker (AIR Base) and a portable microphone (AIR Satellite).
The AIR Base is a smart speaker with a powerful CPU to run voice apps. It is like an Amazon Echo device, but with local processing of the user’s voice. It comes packed with features, such as an HD speaker, a far-field microphone array and a powerful CPU. It is also fully customizable, running Debian as the underlying operating system. The main purpose of the AIR Base is to do the heavy processing of voice queries, so that less powerful AIR devices can just forward their queries to it.
The AIR Satellite on the other hand is a battery-powered, MCU-based portable microphone. It enables you to control your devices with voice, wherever you are in your home.
The AIR Satellite is also tactile, enabling multimodal interactions such as dimming the lights by swiping up and down on it, or play/pausing the music by tapping it. The tactile interface is fully programmable, and acts as an extension of the various Snips voice apps.
Finally, and importantly, the AIR Satellite is context-aware: depending on the situation of the user, it changes behavior. For instance, when controlling lights, the AIR Satellite will recognize the room the user is in, and adapt to the devices in it. The user can also ask to switch context simply by asking Snips “switch to XXX app”.
Snips AIR will come with a number of built-in apps for most common use cases, trained using our patented data generation technology and tested by the community.
On top of built-in apps, Snips enables anyone to create custom voice app, making the assistant a actual platform. In our next posts, we will detail how developers can create, publish and monetize their apps, as well as how users can discover, purchase and use new apps. Both our developer tools and our consumer AppStore will run on top of an application-specific blockchain, enabling both privacy and incentivization of the community.
Why we care so much about Privacy
Privacy has been a core principle of our company from the day it was founded. Aside from the obvious ethical side of it, we have a thesis that demand for privacy grows proportionally with the number of connected devices we use.
Indeed, back in the early days of the internet, digital Privacy meant protecting what you did online. This wasn’t as much of a big deal since most of our life was spent outside our browsers, doing offline things that weren’t tracked. But as we started using smartphones and connected devices, digitizing every aspect of our lives, the line between online and offline started to blur. And the more we get connected, the more our personal data gets recorded and analyzed, opening the door to Privacy breaches that extend beyond The Internet and onto every aspect of our lives.
This is why for us at Snips, it is absolutely obvious that putting a microphone in your most private space — your home — must be done with Privacy by Design. No one should ever be able to listen to you, surveil you, manipulate you or hack you. Because what happens in your home, should stay in your home.
Read more about the privacy issues of cloud-based voice assistants
Read the French Privacy Authority report on voice assistants
If you liked this article and care about Privacy, smash that clap button, then tweet everyone 👉👉 randhindi & snips
Interested in understanding what’s going on in tech these days? Follow my instagram stories @randhindi 🤓
Interested in getting in early on the Snips token sale? Visit token.snips.ai or join our telegram group t.me/snipsair 💸
