Edge AI voice processing reference design

1 min read

Syntiant, a provider of deep learning solutions, has developed a new reference design that will enable customers to deploy ultra-low-power, edge AI voice processing applications in true wireless earbuds.

Incorporating the Syntiant NDP100, the ultra-low-power wireless earbud reference design enables always-on AI voice processing at under 140 microwatts, delivering 100x more efficiency and 10x the throughput compared to traditional MCU-based systems. With custom wake word support, as well as compatibility with other voice assistant ecosystems and in multiple languages including Mandarin and Cantonese, the NDP100 enables hands-free control of local speech commands, such as playing music or operating a mobile phone.

“We continue to see an increase in demand for voice functionality as people rather use hands-free speech to control their devices,” said Kurt Busch, CEO of Syntiant. “Our edge AI reference design brings a highly accurate voice interface to earbuds that is more efficient and consumes minimal power, while also providing the data, machine learning training and support, giving manufacturers a complete end-to-end solution for easy deployment in any country and in any language.”

Syntiant’s edge-optimised training pipeline enables ODMs to quickly build neural networks for a variety of always-on voice, sensor and image applications, providing customers with low-cost, turn-key solutions that put deep learning processing into almost any edge device, free from cloud connectivity, ensuring privacy and security.

The company’s data platform has been designed to ingest data from crowdsourced or proprietary sources, expand the data set with synthetic data generation and produce high quality datasets for training.