When we started Snips in 2013, we were a research lab in Artificial Intelligence. Pretty quickly, we realized that the biggest problem facing our generation was the way humans and machines interact.
Not Man vs. Machine — Man + Machine
It’s us humans who need to learn how to use all these machines. This means the more technology we want to use, the more stress we are putting on our brains. The result? We develop coping mechanisms: we either build habits and become robots, or we flat out refuse to use more technology.
Fortunately, we have a solution: Artificial Intelligence. By making machines understand and speak human languages, we can ask them what we want instead of learning how to use them. And since language is a universal interface that can be used to express anything, this would democratize access to technology (my dad can speak better than he can use his smartphone!) and make it so easy and intuitive to use our devices that we simply won’t have to think about it.
Just like electricity before it, AI will become ubiquitous and make technology disappear from our consciousness.
This is why we started Snips —to put an AI assistant in every device and make technology disappear.
The Snips Voice Platform
When we raised our Seed round in 2015, we had a simple plan: build technology, test it in products, then make it available to everyone. Today, we are proud to announce we reached our goal, and are launching our voice platform for connected devices 🚀.
It’s an end-to-end solution that enables anyone to add voice to their product. Whether you are building a car, a TV, a music speaker or just hacking on a Raspberry Pi, you can use Snips to create a custom voice assistant and make your users interact with your product via voice.
Our platform comes with many features: customizable Hotwords, Speech Recognition, Natural Language Understanding and Dialog. The only thing we don’t do yet is Speech Synthesis, but that will come soon 😉.
Our platform uses the latest advances in Deep Learning to provide the highest performing voice solution on the market. A recent benchmark shows that we largely outperform every major platform at Natural Language Understanding tasks for custom intent creation, including Amazon Alexa, Google API.ai, Facebook Wit.ai and Microsoft Luis.ai.
I’m really proud that our team of 40 people in Paris and NY managed to get such high performances when there are hundreds of people working on the other platforms.
The magic behind our platform is something we call Training Set Synthesis (TSS).
In a nutshell, the idea is to generate artificial training data to train machine learning models, instead of relying on real user data. By combining machine learning with human operators, we are able to create tens of thousands of high quality examples of what real users would ask the assistants, at the touch of a button. This solves the cold start problem, as it means we can train our models with thousands of examples instead of the dozens people usually write by hand.
We were also able to generate a state of the art speech recognition engine, performing similarly to Google Cloud Speech. But since our NLU performs best, our ASR+NLU stack ends up performing better than using Google Cloud Speech + API.ai.
As you might have noticed from the legend in the image above, not only does Snips perform better than API.ai, we are also doing everything COMPLETELY ON DEVICE, with absolutely nothing happening in the cloud! You can have your voice assistant run directly in your device or smart hub, without the need for a backend infrastructure. We pushed the limits of Deep Learning so far that we can now do real-time ASR and sub-100ms NLU on a Raspberry Pi 💪.
Why does this matter? Because of Privacy!
It is important to remember that voice is biometric, meaning it can identify you uniquely without anyway to change your identity. You simply can’t reset your voice like you reset a cookie in your browser. Just like we shouldn’t have your fingerprints, it is pretty obvious that we shouldn’t have your voice. By making the AI run on the device that people interact with instead of in the cloud, we are able to guarantee their Privacy by Design.
This is not only the right thing to do ethically, it is also the only way to comply with the upcoming European regulation on Privacy, the GDPR. Frankly, any company who wants to do business in Europe after May 2018 will have to follow Privacy by Design principles. It doesn’t matter if it’s an American, European or Chinese company; if they want to offer a product in Europe, they will need to offer Privacy. As the only Private by Design Voice platform in the market, we are also the only solution to guarantee GDPR compliance. Use cloud-based voice providers at your own risk!
We often hear that to make better AI, we need to get more user data, thereby compromising on Privacy. This is absolutely wrong.
Remember: we can actually generate training data instead of relying on real users! Not only are we able to get better performances than providers tracking millions of people, we are also able add new languages easily. Our platform is already capable of understanding English, French, German, Spanish and Korean, and we will be able to understand a dozen more within a few months.
Our entire technology is accessible through a simple web interface, where voice assistants can be created in just a few steps. Right now, it can only be deployed on a Raspberry Pi, but more platforms are coming soon. And it is completely free for makers and for prototyping!
A $13M Series A round
Reclaiming digital territory from the GAFAs won’t be an easy task. This is why we are proud to announce we closed a $13M Series A financing round with 3 amazing investors: K-Fund 1 managed by Korelya Capital, PSIM managed by BPI France and MAIF Avenir. Eniac Ventures, who invested in our Seed round, is also participating.
This fresh funding will enable us to keep pushing the limits of AI and Privacy, adding more features to power Voice assistants for every device on the planet. It will also enable us to grow our team in Paris and New York, both on the sales side and technical side.
This is an important milestone for us because for years we were told nobody cared about Privacy. We were told that data was the name of the game. Wrong.
Through grit and focus, we were able to solve the AI + Privacy paradox, shipping a voice platform that is both powerful and Private by Design.
Let’s make technology disappear!
If you believe in AI + Privacy as much as we do — find us on Product Hunt!
If you enjoyed this article, it would really help if you hit recommend below :)
If you want to work on AI + Privacy, check our jobs page!