Keen Research Brings the Power of Speech Recognition to AR

To most people, the term “Enterprise Augmented Reality” conjures up images of headset-wearing technicians interacting with complex machinery guided by overlaid imagery and text instructions.

But anyone who thinks of AR in those terms alone is missing an important component that can greatly increase the value and productivity of an AR solution, according to Ognjen “Ogi” Todic of new AREA member Keen Research. That component is speech recognition.

“Imagine a field service technician repairing an industrial machine,” said Todic. “Would that technician rather spend time swiping through a thick virtual service manual to find the section they need, or simply say ‘drive assembly replacement’ and instantly get the necessary instructions?”

That example highlights the power of Keen Research’s KeenASR SDK, a toolkit that enables AR solutions providers to leverage the power of speech recognition in their software.

The KeenASR SDK makes speech recognition practical for AR use cases because all processing is done on the device. That means it can work when users are beyond network reach, there are no latency issues, and it avoids the security and compliance concerns that arise when processing is performed on the backend and private data is sent over a network. Using speech recognition is also a must for use cases that require workers to have both hands available to perform their tasks.

Founded by the Serbian-born Todic in 2004, Marin, California-based Keen Research began as a service provider focused on software development. In 2016, the company sharpened its focus to concentrate on building and licensing its SDK. Since then, its offering has been integrated in a number of mobile applications across many verticals.

Keen has made inroads in a variety of industries where speech recognition delivers value. The AmbiFi SaaS platform leverages KeenASR for its virtual guides, smart procedures and protocols, checklists, and workflow. eoStar has included KeenASR in a voice picking application for a Warehouse Management Solution. And Ascentis has made speech recognition a core feature of its touch-free time clock solution.

For AR solutions, the KeenASR SDK enables organizations to include hands-free voice control, hands-free data entry, domain-specific virtual assistants, and domain-specific dictation/notes, which are particularly useful for remote assistance and training. Its ability to work in noisy environments is an added plus.

In addition to voice-enabling AR headsets and software running on those devices, Keen Research sees a great opportunity in audio/speech-based Augmented Reality. “For most people, Augmented Reality means augmented vision; audio and voice are somewhat neglected. But there are a number of use cases where audio/voice-only experiences can be effective,” said Todic.

KeenASR SDK is designed to work effectively across a broad range of operating environments and devices. It works on mobile devices running iOS, Android, and Linux, as well as standalone AR and VR headsets. It requires approximately 100MB of RAM and 30-40MB of on-device storage. For those seeking further assistance, Keen Research offers professional services for SDK integration, proof-of-concept development, customization of language and acoustic models, and porting to custom hardware platforms.

Ogi Todic sees his company’s AREA membership as an opportunity to both learn and teach.

“Within the AREA, we can learn about the needs of Enterprise AR users directly from companies implementing AR solutions, which is important to ongoing development of our SDK as well as new products,” he said. “In addition, we look forward to showing AR solution providers how speech recognition can help them deliver solutions that fulfill a broader range of needs and offer an even greater return on investment.”