Well known as the leading supplier of MEMS microphones Knowles moved into the adjacent field of audio DSP through the acquisition of Audience Inc. in 2015.
The IA8201 is designed for advanced audio and machine learning applications, enhancing power-efficient intelligence and privacy at the edge. It offers voice activation and four channels of microphone audio processing optimized for power-sensitive applications with the compute power to perform advanced audio output, context awareness and gesture control for consumer electronics.
The IA8201 includes a 128-bit core (DMX) with Knowles proprietary instruction set and a Tensilica HiFi3 core (HMD), both with Knowles audio enhancements. The DMX is a 4-way floating-point SIMD processor targeted towards efficient performance computing (e.g. beam-forming, barge-in, AEC), while the HMD is targeted towards efficient, low-power, wake-on-voice applications with a two-way floating-point SIMD processor. Both cores contain dedicated accelerators for FFT, peak finding and DNNs.
A broad set of audio, and general purpose interfaces enable flexible interfacing with digital microphones, other sensors, and a host for further processing. 1.44Mbytes of user RAM enables storage of multiple algorithms and voice keywords.
“Today’s consumers expect better listening accuracy in their electronics, which drives systems that have more mics and more processing. Device makers are also highly motivated to enhance user privacy and keep more personal information on-device. Together, this creates huge challenges for designers to implement these enhanced capabilities locally versus in the cloud,” said
“Knowles AISonic Audio Edge Processors provide advanced multi mic processing at 10 to 100 times the efficiency of generic processors, which helps our customers design compelling products with performance and privacy features that end users demand,” said Mike Polacek, president of intelligent audio at Knowles.
The IA8201 audio processor packs 1.44Mbytes of memory and dual 175MHz cores with voltage scaling in a 2.6mm by 3mm eWLB package and a 6.0mm by 6.0mm QFN package.
Interface options include two each of I2C, SPI, UART, and 24 GPIOs with interrupts to support various sensor and data connections, plus 4x PDM in, 2xPDM out, 3x I2S/TDM (4ch in/out) to support up to 4 microphones and various audio connections.
Knowles’ open DSP platform facilitates partnerships with third-party algorithm providers who will be rolling out IA8201-compatible keyword models, various context detection plugins, and machine learning platforms including TensorFlowLite. The ROME framework which forms the foundation of the operating system is open to developers.
The IA8201 audio processor is sampling now to customers and will be ready for mass production in 3Q19.
Related links and articles: