Syntiant and Sensory Partner on Ultra-Low-Power Multilanguage Voice Solution at the Edge
Syntiant, a semiconductor start-up providing intelligent voice solutions at the edge, is partnering with Sensory to deliver a deep learning, multilingual voice interface for battery-powered devices.
The combined solution merges Sensory's TrulyHandsfree wake word engine and voice control with Syntiant's microwatt-power neural decision processors (NDPs), bringing low-latency, real-time inference to edge devices and enabling consumers to voice commands in dozens of languages. Working together, the two technologies could also support additional features, such as voice-based user identification.
"For a busy mom in Korea setting a house alarm or teenager in Barcelona raising the volume on his smart speaker, voice commands are becoming more ubiquitous driven by worldwide consumer demand," said Kurt Busch, CEO of Syntiant, in a statement. "Collaborating with Sensory allows us to combine their AI with our silicon technology, providing customers a large multilanguage library of local commands for just about any application."
Syntiant and Sensory are accelerating the delivery of a multi-language interface in devices, such as earbuds, smart speakers, and smartphones, at low power. Custom built to run neural workloads, the Syntiant NDP100 and NDP101 can support dozens of local voice commands and consume less than 140 microwatts while performing local processing of audio events. In addition to voice triggers, other device capabilities include audio event and environment classification, as well as sensor analytics.
"Sensory's embedded AI voice software combined with Syntiant’s neural network technology opens up a new market for both companies, bringing ultra-low-power, multilanguage, local command support to the edge," said Todd Mozer, Sensory's CEO, in a statement. "Our joint solution provides added security because commands are performed at the edge, free from a cloud connection, while devices and applications come alive without pushing a button."
VoiceHub's interface allows users to build vocabularies in dozens of languages without coding experience.