In recent years, MEMS technology has made a remarkable impact on acoustic applications, enabling the creation of cost-effective compact audio systems. These systems serve a variety of cutting-edge applications, including consumer electronics, medical devices, automotive systems, and more.

MEMS microphones are extensively used in mobile phones and wearable devices to deliver high-quality audio for calls and recordings. In the automotive and industrial sectors, they enable hands-free calling, voice control, and even vibration monitoring for predictive maintenance. In medicine, they serve critical functions in devices like smart stethoscopes, blood pressure monitors, and abnormal heartbeat detection systems. These microphones are now vital to the Internet of Things (IoT) ecosystem.

But with microphone models’ competition primarily about price, what factor could truly drive a breakthrough in this space?

A good answer came late last week as Knowles Corporation agreed to sell its consumer MEMS microphones business to Syntiant. Syntiant, which focuses on neural network hardware and software models, acquired this business to bring microphones to the next level by integrating AI.

MEMS microphones are integral to smart voice assistants like Amazon Alexa and Google Home, enabling them to operate based on user voice commands. They are also part of smart remote controls for TVs and air conditioners. Additionally, these microphones have potential applications for Extended Reality (XR) devices, facilitating communication through headsets.

How would all these voice management systems benefit from AI? Voice activity detection (VAD) improves total audio system power consumption, allowing it to remain in sleep mode if a voice is absent. Another task neural networks perform very well is voice extraction (VE).  VE is crucial as True Wireless Stereo (TWS) devices gain momentum due to the over-the-counter (OTC) hearing aid boom.  Large and small vendors consider VE the next bestseller, given its potential to enhance the audio experience across various user scenarios.

Neural networks solve voice processing tasks very efficiently. The question then is what power consumption the processors that run the neural networks need. POLYN believes that neuromorphic analog processors are the best answer, thanks to their minimal power usage.