On the “Edge” of AI Engine Processing

by | Friday 29th March 2019

Capitalizing on the AI Ecosystem

This month we focus on the expansion and acceleration of AI use case applications.  There has been significant growth in the smart speaker and voice assistant market. The global smart market is expected to reach 96 million units in 2019, growing to 158 million in unit sales by 2023, according to SAR Insight & Consulting.

A number of connected devices from smart speakers and headphones to automotive are taking advantage of the growing AI ecosystem, and delivering smart functionality with the support of dedicated and powerful processors. These new AI engines and enabling technologies are driving the latest in innovative and immersive solutions.

Voice Processing Migrating to the Edge

As part of the AI engine, a number of integrated chip, software, and sensor-based companies are contributing and providing important pieces of the audio and voice signal processing applications.   Several new products have been announced to bring AI services like voice triggers, speech recognition, natural language processing and voice biometrics to the network edge for a variety of use cases.  A number of companies have recently released smart audio and voice solutions such as Qualcomm, Cirrus Logic, MediaTek, DSP Concepts, CEVA, DSP Group, Sensory, Vesper, and Synaptics.  These companies are now developing high-performance, low-power solutions that are suitable for embedded and more secure AI edge computing.

Examples of recently announced solutions include complete integrated audio processing SoCs (such as the Qualcomm QCS400 Series) that combine enhanced audio, voice command and expanded connectivity to link various connected devices throughout the smart home, Smart MEMS microphones (such as the Knowles IA610 SmartMic and Vesper VM2020) that are or will be embedded in a growing share of smart home appliances, chip technology that allows for voice biometrics at the edge (such as Cirrus Logic and their growth in edge signal processing for secure user authentication)  and other local device processing for independent speaker verification to identify the speaker without the need for a specific wake word (such as Sensory TrulyHandsfree 6.0).  A significant amount of engineering effort is being placed on integrating custom silicon specifically developed for processing artificial intelligence to support an expansive number of smart devices.

Processing for Performance

All of these smart devices that leverage machine learning algorithms are expanding the AI use case and driving local processing – resulting in faster performance and response, lower latency, power efficiency, cost savings, and improved security. Now a range of voice enhancement and UI technologies inclusive of voice triggers, echo cancellation, automatic speech recognition, and natural language understanding will be able to migrate over time from cloud to on-device processing.

SAR Insight & Consulting is producing a FOCUS report covering voice processing implementation. The report will size and forecast edge versus cloud and hybrid computing by device and voice technology.


Contact the author:

Dennis Goldenson