Author
Jerome Schang
Jerome is Head of Global Cloud Partnerships at NXP and fosters leading partnerships for NXP's edge computing and machine learning solutions.
With voice command rapidly becoming an indispensable part of navigating every aspect of daily life, NXP recognized early on the important role that edge computing plays in addressing the growing demand for voice detection across a wide variety of smart home, smart office, industrial 4.0 and smart retail products.
From high performance microprocessors and cost-effective microcontrollers to enabling flexible implementations with completely turnkey solutions, NXP is a proven leader in providing a broad range of scalable options for adding voice control directly into any device. And since all voice solutions must communicate with the cloud, our latest collaboration with Microsoft extends intelligent cloud computing to the intelligent edge: a powerhouse combination that changes the paradigm for advanced voice detection capabilities.
In the past, voice solutions have not only relied on their ability to communicate with the cloud, but have assumed users were always connected. These solutions have also required extensive wake word training with heavily customized modeling of human utterances that would ultimately be deployed in the cloud.
In this paradigm, cloud-based platforms have certainly had sufficient computing power to perform sophisticated natural language processing that figures out what the user means, even if he or she says it in many different ways. Embedded processors on the edge device, however, simply detect speech (wake words) intended for that specific device to avoid obvious privacy and bandwidth issues that would arise from continuously streaming audio data to the cloud.
As a result, wake word training has been a crucial component to ensuring overall efficiency, user privacy and accuracy for speech recognition, but the training process has typically taken a lot of time, storage capacity and manpower to achieve.
NXP’s advanced offline machine learning capability and embedded processing combined with Microsoft’s cloud expertise is changing all of that. Not only does the collaboration enable autonomous computing at the edge to eliminate the need for cloud connectivity, it also dramatically simplifies the wake word training. Instead of requiring months of standard wake word training, combined solutions from NXP and Microsoft can accomplish this in simply a matter of hours. All trainings can take place in the cloud, be deployed at the edge and run independently at the edge on an NXP Arm®-based device. Once training has been completed, a model can be deployed without any reliance on the cloud. This embeds more resiliency into the solution while adding a layer of inherent privacy since voice samples never leave the edge device (nor will they be stored or analyzed in the cloud).
As a result, NXP and Microsoft deliver a powerful, end-to-end solution that addresses four key requirements for any voice command solution in the market today:
Extending intelligence from the cloud to the edge now enables a broad range of new and robust computing, including voice detection and many other use cases, where the intelligent edge becomes the new, privacy-preserving arena. Join us CES 2020 for live demonstrations that showcase how NXP and Microsoft are forging the future of voice command solutions for a smarter connected world.
WHERE: NXP Booth, CP-18
WHEN: CES Conference Dates: January 7-10, 2020
Head of Global Cloud Partnerships, NXP Semiconductors
Jerome is Head of Global Cloud Partnerships at NXP and fosters leading partnerships for NXP's edge computing and machine learning solutions.