IA8201 Audio Edge Processor enables accelerated machine learning inferencing

20 February, 2022

1.172

Views0 Comments

Knowles’ AISonic™ audio edge processor IA8201 audiocentric OpenDSP enables accelerated machine learning inferencing

The Knowles AISonic audio edge processor IA8201 is a high-performance, ultra-low-power audio-centric OpenDSP supporting up to 4 mics, multiple high-speed interfaces, and GPIOs in two package options (eWLB and QFN). It provides low power, high efficiency, privacy, and compute power enabling customers to design modern products with far-field voice processing functionality for accurate listening. The IA8201 enables wake-on-voice processing for low latency voice UI, noise reduction, context awareness, and accelerated machine learning inferencing for edge processing of sensor inputs.

Knowles currently utilizes a modular approach to add voice with the capability to work with existing platforms and MCUs. IA8201 allows easy integration with legacy MCU system boards by running these commands over a simple UART interface.

The IA8201-RDI-01 is an evaluation kit based on the Knowles AISonic audio edge processor IA8201. It is a complete system-level solution that enables Knowles customers to quickly go from an idea to a front-end audio solution that meets their platform requirements. It includes an IA8201-LT dev board, Raspberry Pi connector board, one 3-mic array, one 2-mic array, and flex cable connectors. The system firmware release to connect this platform is available on solution.knowles.com. It also comes with support of Sensory Voice Hub, which allows a user to create wake-words and local commands using a Web-based voice user interface.

Use Case Examples

Low power voice wake: Listens for specific OEM keywords to wake the host processor. Large memory enables processing of multiple stages on-chip for accurate results.
Proximity detection: When combined with an ultrasonic capable speaker and microphone, detects the distance between the system and an object; can replace an IR-Prox sensor in bezel-less phones.
Hub: Determines location of voice source while tuning out a noisy environment and lowering music to detect voice commands. Simultaneously takes metadata input and overrides beamformer to focus on camera-tracked objects.
Security system: Activate with a voice command. Detect glass breakage/smoke alarm, log direction of noise source, trigger alarm, and send alerts through Wi-Fi connection.
Wireless earbuds: Delivers low power premium wake-on-voice performance, talk detection to eliminate false triggers, enhanced voice quality through advanced beamforming and noise reduction algorithms, and support for local commands, including answer/ignore calls.

Features

Multi-Core: DeltaMax, optimized for compute; HemiDelta, optimized for low power
Audio interfaces: Up to 4x PDM digital microphones – 1 stereo inputs, 4 x mono inputs, and 1 stereo output, supporting clock rates up to 6.144 MHz; up to 3x PS/TDM ports supporting 8 channels each of 32-bit audio data using a 24.576 MHz input clock
Control interfaces: SPI, I²C, UART, available GPIOs
Memory: 1.44 MB RAM (1 MB available to users)
Clock: 175 MHz
Packaging options: eWLB 3.00 mm x 2.6 mm x 0.715 mm, 0.4 pitch, 42 ball; QFN 6.00 mm x 6.00 mm x 0.75 mm, 0.5 pitch, 40 lead
System requirements: IA8201BC 1.8 V_DD, IA8201CQ 1.8 V_DD and 3.3 V_DD, -20°C to +85°C