Snips has grown by leaps and bounds over the past year. Meet the engineers, machine learning experts, and international business leaders working to bring groundbreaking edge computing and voice technologies to the market.
Snips was founded by Dr. Rand Hindi, Dr. Maël Primet, and Michael Fester in 2013 as a research lab in Artificial Intelligence. Since then, we’ve evolved as a company to offer an AI-powered voice platform designed for connected devices, which gives firms the ability to add customized voice recognition to their services and goods. We now sell our voice technology as a white label solution for enterprise device manufacturers. With over 40 enterprise customers across the smart home, building, entertainment, retail and professional services, and industry 4.0 sectors, and dozens of partners from chip makers to hardware integrators.
Our voice platform is free for non-commercial use, with a web interface that empowers makers and developers to create their own assistant. With a developer/maker community that is actively engaged in building voice assistants, Snips has created one of the fastest growing voice app developer communities.
Voice makes our interactions with devices so seamless that eventually we feel that technology has disappeared.— Dr. Rand Hindi, Chief Executive Officer
At Snips, we believe that data privacy shouldn’t be an afterthought, but rather at the core of every technology product we use. We’re proud to have built the first private-by-design voice assistant. Showing that it is possible to give users the best in AI while processing voice data locally only. Even more, Snips now holds the Trustable Technology Mark, a first-of-its-kind certification for IoT companies that puts our values into action as we lead the way in trustable technology and user rights.
Our Team DNA
Last year, LinkedIn highlighted young upstarts at the forefront of innovation in France, with Snips named the 2nd best start up to work for in France. We were also chosen by Wired as one of the hottest startups in Paris.
Today we are a team of more than 78 people based in Paris, New York, and Tokyo. All of our R&D efforts are located in Paris. More than 60% of Snips employees are engineers, 15% hold PhDs, and the company has filed 8 patents for different bricks of its technology. We always encourage female engineers and developers to join the company . Currently women account for 25% of our workforce and we are on-track to improve this ratio in the coming months.
Our Technology: cloud-level accuracy on the edge
Research & development is an integral part of the Snips DNA. Furthermore, the machine learning team is committed to contributing to the spoken language understanding community through academic publications. Alice Coucke, senior machine learning scientist at Snips, has been busy co-authoring 4 research publications in the past year. With a PhD in Statistical Physics from Ecole Normale Supérieure, she joined Snips in early 2017 and has bounced around different teams since, to lend her expertise in language modeling and keyword spotting.
Along with fellow Snipsters Alaa Saade and Alexandre Caulier, she recently undertook a far-reaching benchmarking experiment. The intention was to demonstrate that the Snips Voice Platform has performance on par with cloud-based commercial solutions, while running on small devices typical of IoT applications. In a recent publication, they outlined the full design of the Snips embedded, private-by-design spoken language understanding system.
Reaching cloud-level performance on such a large vocabulary has been a key milestone for the Snips team. But we do not intend to stop there: further improvements will come with deeper use of the context. For example through speaker recognition, cleaner separation between a user’s voice and background noise, and by dynamically adapting to the user’s vocabulary.
The Snips team has reached a milestone: cloud-level Spoken Language Understanding is now possible on the edge. This revolution will deeply transform the way we interact with technology. Opening the way to a reliable, personal, and private Artificial Intelligence.— Joseph Dureau, Chief Technology Officer
In addition to this work, the Snips team celebrated three other research publications being accepted for publication to the workshop on privacy in machine learning and artificial intelligence at ICML 2018 (International Conference on Machine Learning). And to the main track of ICASSP 2019 (International Conference on Acoustics, Speech, and Signal Processing):
- “Federated learning for keyword spotting”
- “Efficient keyword spotting using dilated convolutions and gating”
- “Snips Voice Platform: an embedded Spoken Language Understanding system for private-by-design voice interfaces”
Always new languages
The language team, which works on the localization of Snips technology to new languages and the development of linguistics tools, welcomed María Isabel “Maï” Díaz Sáez in September 2018 as a language expert for Spanish. With her Master’s in Translation and Interpreting from Universidad de Salamanca, Maï worked closely with Odile de Vismes and Rosa Stern, respectively Senior Linguistic Resources Manager and Senior Computational Linguist, to help build and test a phonological model of Spanish and create voice apps in Spanish for the Snips app store. Since launching Spanish as a supported language, Maï has assumed a localization specialist role. Working with David Leroy, Principal Machine Learning Engineer at Snips, on cross-language support for a multi-domain, large vocabulary assistant.
With a focus on ontology definition, language localization, and quality assurance, the team has successfully launched 6 supported languages for the Snips platform . French, English, German, Japanese, Spanish, and Italian . And is on track to an even greater goal: supporting languages spoken by 50% of the global population by 2020.
A New Suite of Edge Products
Coming out of R&D last year, Snips became the first provider of embedded voice technology for OEMs that provides both full support for micro-controllers and full natural language support for micro-processors. The product team has made major progress towards collaborating productively with enterprise-grade partners and introducing a new suite of edge products.
The MCU team at Snips was created to bring our voice assistant to even smaller, cheaper devices. The team is led by Hubert de La Jonquière. He joined in 2015, after studying General Engineering at Ecole Centrale Paris and Computer Science at Imperial College London. Since its creation 8 months ago, the MCU team continuously optimizes voice recognition algorithms and works closely with semiconductor companies including NXP. The team works on running state-of-the-art deep learning models to enable a large array of wake words and voice commands for low-powered devices, such as smart appliances. This new technology will now be distributed in partnership with DSP Group, through a new product line called Snips Satellite, bringing voice recognition capabilities to millions of devices in 2019.
The engineering and product teams at Snips have delivered products that nobody could have imagined two years ago. They brought voice recognition and natural language understanding formerly running on an army of servers in the cloud to computers no bigger than a few square inches (or several square centimeters).— Sébastien Maury, Chief Product Officer
The Snips engineering team programs primarily in Rust. Snips Principal Engineer Mathieu “Kali” Poumeyrol used Rust at a previous job, writing multi-platform code. He pushed hard for Snips to consider adopting Rust. As a programming language, Rust now helps the engineering team achieve efficiency, portability, and safety . Also, it has the performance characteristics to be able to run very fast, even on small devices. Two years ago, in an attempt to mutualize some code between Android and iOS, Thibaut “Fredszaq” Lorrain wrote a few lines of Rust. That eventually became Snips Flow.
As an embedded natural language voice recognition solution that runs on-device, Snips Flow achieves performance equal to or better than cloud-based solutions. Fredszaq, who graduated from the National Institute of Applied Sciences (INSA) in Rouen and joined Snips in 2016, is part of the device team engineering the future of Snips Flow. His team works to make Snips Flow available on Raspberry Pi, Android, iOS, and other small CPU-based hardware platforms. The device team designs the overall architecture of the different components of Snips Flow. As well as the communication between these components and third-party apps.
The team is continuously bringing in fresh talent. Noé Baylac joined Snips in February as an intern to help consolidate the current components of Snips Flow and develop new features. As a master’s student at the Université of Pau, his primary goal is to learn the ropes of embedded engineering, while working on a real product. At Snips he is undertaking learning in situ. Benefiting from the range of Snips products that raise interesting computer science problems and a work environment that facilitates collaboration between novice and seasoned engineers. During his first month, Noé contributed to key components of Snips Flow. Improving the product architecture, while also working on one of the company’s open source tool.
With Snips Satellite, Snips Flow, and Snips Commands now in market, we’ve started 2019 with recognition from US and international technology conferences. Winning the CES 2019 Best of Innovation Award in the “Embedded Technology” category and the “Best in Show” Award from Embedded World 2019.
A Strong Network of Clients and Partners
In just the short span of 15 months of commercializing its award winning voice technology, the Snips Business Team led by Stephen Bauer (VP Sales) has connected with over 1000 top tier companies. Among which, 40 — and counting! — have already worked with Snips to bring voice interfaces to the market. Furthermore, their business operations varies within different uses cases including smart home, smart building, entertainment, retail, financial services, industrial maintenance and supply chain. Snips comprehensive product suite and easy-to-use tool chain makes it easy and flexible for manufacturers and organisations to design and leverage voice interfaces tailored to their products and services. It can run entirely offline without the normal sacrifices to brand hijacking or requirements for cloud connectivity that come with traditional solutions. Over the course of 2019, Snips will launch in new voice-controlled devices in a broad variety of markets.
We have reached market resonance. Indeed, almost every single product or service company is now gearing up for voice enablement. Snips has become a much needed alternative to mainstream voice assistants by offering strong privacy principles, low footprint, and industrial strength on-device performance.— Yann Lechelle, Chief Operating Officer
The Snips Partner Program was officially launched in February 2019. To cultivate and grow an ecosystem of leading Technology and Solutions services that enable OEMs to integrate and go to market with voice faster and stronger. The Snips Partner Program gives access to high-performance hardware, specialized voice interface design, audio component design, parts selection and end-to-end software integration support.
Together with this partner ecosystem, Snips is filling the gaps in integration between our voice recognition software and client products to help OEMs and service providers bring their voice-enabled products to the market faster. Partnerships also already ongoing include NXP, Crank and DSP Group on the technology side and Smile on the solutions side.
A Fierce Developer Community
The Snips platform empowers developers and makers to build Private by Design and on-device voice assistants for their projects and also to share their voice apps with the community. Our tools are free for developers who want to tinker with our technology for non-commercial use. Also, many of our tools are now open source, creating endless possibilities for what can be built. We currently have a community of over 23,000 developers and makers (and counting) since launching the web console in mid-2017.
Together with partner Seeed Studio, we launched two Voice Interaction Development Kits. These are available for international purchase as a Base and Satellite integration. This collaboration represents the second iteration of maker kits, an all-in-one hardware + software device devised by Snips. With the goal of making building voice applications and deploying assistants easy for makers and developers. The Base Kit runs the full Snips AI Voice Platform on a Raspberry Pi 3 B+ and the Satellite Kit extension adds voice control to several rooms with a Raspberry Pi Zero W.
Snips strongly believes in the importance of open source to improve transparency and trust. This empowers the community to participate in the development of new features and pushes AI furthermore to the edge. Snips made the decision to fully open source our natural language understanding (NLU) engine to contribute to the development of AI and to reduce the dependency on centralized providers. The Natural Language Understanding engine under the Apache 2.0 license is an open-source, on-device, private-by-design alternative to Dialogflow, Amazon Lex, and other NLU cloud services.
Putting an AI in Every Device
Every day Snips gets closer to its mission of making technology disappear. The team has been working to achieve a future where technology is hyperconnected and advanced. But yet so seamlessly integrated into our lives that we never have to feel that it exists at all. With a team now doubled in size, expansion into Tokyo, additional language support, and new products hitting the market in 2019, Snips is certainly well on its way towards the ubiquitous personal voice assistant and putting an AI in every device.