On the Edge: TinyML and Voice Recognition Technology Are Transforming Industries

Rising demand for portable, light footprint voice recognition devices is fueling a growth of the tiny machine learning (TinyML) field. Offering a broader, high-level discussion specifically focused on voice recognition, Vikrant Singh Tomar, the founder and Chief Technology Officer of Fluent.ai, explains the impact for embedded voice and the overall audio industry. This article was originally published in audioXpress, February 2022.

Aspinity launched the first member of its analogML family, the AML100, which is the industry’s first and only tiny machine learning (ML) solution operating completely within the analog domain. As such, the AML100 reduces always-on system power by 95%, allowing manufacturers to dramatically extend the battery life of devices and migrate walled powered always-on devices to battery - opening whole new classes of products for voice-first and smart home systems.

The February 2022 edition of audioXpress is now available and features our annual Market Update on Microphone technology. Vikrant Singh Tomar discusses TinyML and Voice Recognition Technology On the Edge, Richard Honeycutt writes about Quantification of Speech Intelligibility, and Mike Klasco expands on audio applications for Haptic Devices. George Ntanavaras builds a phono and line/headphone amplifier project, Brian Bylenok proposes a refined Current Feedback Headphone Amplifier, and Richard Honeycutt reminds us that Tube Power Amps also have Classes. Now available in print and online.

Syntiant unveiled its TinyML Development Board, an easy-to-use developer kit aimed at both technical and non-technical users for building machine learning-powered applications in smart products, such as speech commands, wake word detection, acoustic event detection and other sensor use cases. The new machine learning development board allows developers to build low-power voice, audio and sensor applications using Edge Impulse’s Embedded ML platform.

CEVA announced that its CEVA-BX DSP cores and WhisPro speech recognition software now also support TensorFlow Lite for Microcontrollers, a production ready, cross-platform framework for deploying tiny machine learning on power-efficient processors in edge devices. WhisPro is CEVA's speech recognition software for voice wake words and custom command models. TensorFlow Lite for Microcontrollers is a solution from Google for implementing machine learning at the edge, already optimized and available for CEVA-BX DSP cores.