Capabilities of a Smart Microphone with Embedded Neuromorphic Front End Chip

Smart microphones can be very effective with neural networks' capability for pre-processing data. The neural network on the chip provides the advantage for voice processing tasks such as voice detection, voice intelligibility, and voice command processing. At the same time, it supports user privacy and increases the security of smart home solutions, personal communication devices, hearing assistance, robotics, etc. The question is how to find an efficient way to implement it.

Analog circuits excel at accurately representing real-world phenomena and handling continuous signals. They outperform digital circuits for many 1D signal raw data pre-processing near the sensor. Analog saves power consumption of the sensor node and reduces data to be processed on digital CPU/MCU for classification and decision.

POLYN offers an application-specific neuromorphic analog IC for always-on data pre-processing at the sensor. Neuromorphic Analog Signal Processing Front End (NASP NFE) features ultra-low power consumption, low latency, and high accuracy. It needs only 35uW of power for always-on voice detection or up to 120uW to extract a word command from a noisy background. It is a new element in sensor node architecture and the logical continuation of the sensor itself.

Neuromorphic processing of voice solves the problem of communication in a noisy environment through voice extraction, which differs from noise cancellation and works well with irregular noises. It also improves hearing when sound is received over the network with noises mixed at the sender side. Another use case is smart voice controls for Smart Home, Smart Warehouse, headphones, and helmets. The solution improves voice intelligibility in voice management systems, providing high accuracy and robustness even in harsh environments with noise interference and no cloud connectivity.

NASP NFE supports easy integration with existing solutions and new designs operating on batteries or energy harvesting, making the whole solution more sustainable. NASP NFE chips are best suited for smart devices in various AIoT applications.

POLYN invites all parties interested in adding AI to voice-sensing devices to test our neuromorphic voice processing models on the simulation tool.

This session is part of our “An Engineer’s Guide to AI Integration” webinar series. Registration/event page here.

Sponsored by: