The differentiator in the company's voice technology - available in English, French, German, Italian, Spanish and Japanese - is that they run voice processing completely offline, which means more privacy and the possibility to design solutions that are not dependent on having an Internet connection. At CES 2019 Snips.ai will be launching a range of new products, including Snips Commands, Snips Flow and Snips Satellite that will make it easy for manufacturers to design speech-enabled devices that can operate offline and to market them faster without sacrificing their own brand identity. According to the company, this new product suite will also enable OEM's to explore the value of voice interfaces, from simple voice commands for remote control all the way to full natural language understanding (NLU).
Snips Commands is a voice interface that allows users to easily recognize a set of commands following a wake word. It can be easily integrated into any device with a Cortex-M7 microcontroller and works without an Internet connection. This is an ideal solution for audio components and home theater systems, allowing bands complete control over which commands are recognized - from the standard commands of "play", "stop", "pause", "next", and "previous" for a music player to preset and operating mode activation available in 66 different languages and dialects. According to the company, Snips Commands require extremely low computing and memory usage, making it an ideal solution for designing affordable devices.
Snips Flow is an embedded natural language voice recognition solution that runs on-device with performance that is equivalent to or better than cloud-based solutions, claims the company. Since it doesn't require an Internet connection, Snips Flow can offer a more robust, and more responsive solution that guarantees total data privacy in homes or offices. Since it runs locally, Snips Flow can be customized completely for end users (address book, musical tastes, denominations of the rooms of the house, etc).
The third software package, named Snips Satellite, is the perfect complement to Snips Flow in a home or office. It can be installed on small devices powered by microcontrollers. These satellites, placed in different rooms, can be activated with a wake word and stream commands to any Snips Flow-enabled device via a local network. The device will be able to understand the meaning and react accordingly. Like all Snips products, no personal data will filter out of the home or office.
Of course, all the voice interface solutions, be it command-based or supporting natural language interactions will be always limited to the scope of equipment control and configuration, even when they can be configured to simulate "intelligent" assistants that guide the user. Until now those solutions have been found by users as slow, unreliable and extremely limited. But Snips claims its technology is able to bring cloud-level performance to the Edge with similar or better performance than existing cloud-based voice personal assistants. "Snips’ voice recognition and natural language software works on -device or on-premise with the same or better performance than the leading cloud-based solutions," they state.
The report shown in the graphic above details a recent performance benchmark for Snips’ Embedded Voice Platform, specialized on each of the voice interface domains, compared to an equivalent cloud-based service. This benchmark highlights two different use cases: a small vocabulary domain like smart lighting, and a large vocabulary domain such as music. In both cases, Snips Voice Platform performance is close to human-transcriber level, and higher than or on par with Google’s cloud-based services.
For the music use case, Snips segmented a test dataset into three ‘Tiers’, spanning different levels of popularity among the 10,000 most popular artists according to a public ranking of Spotify charts. The first Tier corresponds to the most popular artists, ranked between 1 and 1,000 (“Tier 1 Artists”). The second Tier is comprized of artists ranked between 4,500 and 5,500 (“Tier 2 Artists”). The third Tier contains artists ranked between 9,000 and 10,000 (“Tier 3 Artists”).
Here’s how the Snips Voice Platform compared to Google Speech-to-Text and professional transcribers on these different Tiers.
According to Snips, this experiment shows that Google Speech-to-Text’s accuracy quickly diminishes with lesser-known artists, while Snips remains highly precise no matter the artists’ popularity. This, because "Snips Voice Platform obeys a radically different, domain-specific logic. It is systematically, and automatically tailored for a use-case, not a general purpose configuration. Specialization ensures a high and consistent level of support across the board, offering better experience on the music use case than Google Speech-to-Text," the company states.
"Snips is the first and only embedded software vendor to be able to cover the entire voice solution chain, from voice commands to NLU - Natural Language Understanding," explains Yann Lechelle, Snips’s COO. “This year has been very successful for Snips. Coming out of R&D last year, we have become the ONLY provider of embedded voice technology for OEMs with full spectrum support from micro-controllers for cost efficient implementations to micro-processors for full natural language support.”
Snips' embedded technology will be shown on Snips’ booth at CES, and will simulate a real Smart Home that will work without Internet connection and protect end-user privacy.