Voice interfaces offer a simpler, safer and more natural way of interacting with devices. The XMOS project has developed an interface that can detect human presence, distinguish one user from another and act appropriately. “XMOS uses a very low-cost microcontroller that runs the voice processing locally, using only a small amount of power and physical space in a device. Additionally, we added sensors to provide local, privacy-sensitive person detection information,” explains Mark Lippett, XMOS CEO. The algorithms developed by the project were incorporated into the XVF3510 product which is already being used by manufacturers, including a European operator for a ‘next generation’ smart home hub.
From sensor, to neural network, to application
Following this product’s release, users requested the ability to customise the voice interface – particularly to incorporate more ‘intelligence’ themselves, rather than relying on third parties. This feedback inspired the project to create an economical development platform for the artificial intelligence of things (AIoT) called xcore.ai. While the platform provides the infrastructure, a software development kit (SDK) allows developers to integrate high-performance, flexible intelligence into their products. The SDK can combine signals from multiple sensors such as microphones, cameras or even small radars and uses a neural network to detect key features, such as a specific word or face. Using a special vector processing unit, normally only found in expensive graphics processors, XMOS has increased the efficiency of this process, while keeping costs low. “To make it easy for designers to use our platform, we offer sensor interfaces and preprocessing functions as part of the toolkit. The neural network uses an industry-standard format so users can build applications quickly with tools they already know,” says Tom Blackie, XMOS project leader. Throughout the project, XMOS engaged with potential users to test designs and gather feedback. “COVID impacted our lab-based prototype testing, so we created an online early access programme, sending ‘Explorer Kits’ to over 30 organisations,” Blackie notes. “Having users testing in real environments garnered fast, concrete feedback on our work.” With privacy and security a high priority, XMOS implemented several techniques, such as encrypting the device’s software. Additionally, data is processed locally, avoiding the need for external, cloud-based services. It is even possible to operate completely free of any network connection.
For better and safer lives
Aside from the public health benefits of contactless control, the XMOS project could improve inclusivity and diversity as voice interfaces remove implicit expectations of mobility, agility and literacy for everyday tasks. Hands-free operations also offer advantages in risky environments. In kitchens for instance, the system’s ability to distinguish between individuals could prevent children from using dangerous appliances, such as ovens. Using a low-power voice interface, instead of a more power-hungry applications processor and cloud services, also offers energy savings, reducing the related carbon impacts. “The AIoT market is forecast to become a USD 3 trillion industry by 2024. We are currently exploring a wide variety of applications across a range of industries, including bio-monitoring opportunities in healthcare. We look forward to driving the evolution of one of the most exciting technological developments of our time,” Lippett underscores.
XMOS, artificial intelligence, intelligence of things, contactless, COVID, hands-free, voice, interface, public health, sensors, security