Knowles Corp. has introduced a new audio processor designed for performance and power in both near and far-field voice applications including mobile phones, ear-worn products and internet of things (IoT) devices, such as digital assistants, smart speakers and TVs.
The IA8505 processor is an open digital signal processor platform that has four heterogeneous cores and 5.7MB of shared memory and is designed for machine learning through deep neural network hardware acceleration, optimized architecture and proprietary instruction set.
The audio processor can perform specific machine learning tasks 10 times more efficiently than comparable processors, Knowles says. And because the device uses lightweight floating point operations, it enables ease of programmability, high dynamic range and low power comparable to fixed point operations.
The four heterogeneous cores are used to help with four different issues related to near and far-field voice applications including performance for both noise and distance, latency for real-time processing such as active noise cancellation and asynchronous sample rate conversion, power consumption and integration of a standard ARM Cortex M4 core for general purpose controller capability.