Sensory’s New VoiceHub Platform Offers Quick Custom Wake Word Creation
Speech technology developer Sensory debuted the new VoiceHub platform on Wednesday as a way for developers to add custom wake words and voice controls to smart devices. VoiceHub provides a simplified pipeline for creating and testing the wake word, with models trained in as little as an hour after submission.
Wake and Test
VoiceHub is aimed at the developers of voice-controlled devices and appliances of all types that choose to skip Amazon Alexa or Google Assistant in favor of a unique option of their own. Sensory’s TrulyHandsfree technology powers the custom wake words and commands of many such devices, more than three billion, according to Sensory, and VoiceHub operates on a similar training system, just more quickly and within narrower constraints. VoiceHub has been undergoing testing for the last few months and is now available for those who request an invitation.
Generally, it can take weeks and cost plenty of money to test new words and vocabulary before sharing a demo. Shortening that process to a few days, let alone an hour, and cutting the cost of paying services for testing could make a huge difference in the final cost of developing an app or device. Voicebot CEO Bret Kinsella tested out VoiceHub in May. As he wrote in the Voice Insider #80 newsletter at the time, “I tried it out last week and had a “voicebot” wake word operating within a demo app within hours. It’s pretty interesting to see how quickly a model can be trained and Voicebot could become the word on everyone’s lips. ;^)”
According to Sensory’s tests of VoiceHub, conducted by Vocalize.ai, VoiceHub’s custom wake words work as well as Alexa or other industry standards and could be used for mass-produced devices in several languages. Using VoiceHub is free at the moment, though going beyond the demo will still involve paying Sensory for its services.
“Sensory applied decades of experience and lessons learned with shallow net technologies to create its highly accurate machine learning models, which are trained with a mix of real and probabilistically-derived synthetic said Sensory CEO, Todd Mozer. “VoiceHub benefits from all of this work and removes any friction related to developing voice UIs for testing purposes. Furthermore, our VoiceHub models set the bar very high for accuracy and overall performance. We are excited to share these tools and capabilities with the speech tech community and beyond and believe VoiceHub will serve as a catalyst for rapidly accelerated innovation of new voice-enabled experiences.”
Sensory has been expanding its offline wake word and voice command features at a steady clip. In June, the company set up a new biometrics platform combining vocal and facial recognition called TrulySecure aimed at helping tech recognize people wearing masks. The quick custom wake word creation is also part of a selfie-taking app Sensory unveiled in June, letting users make any words they want as the voice command for taking a photo.
To make VoiceHub work, Sensory partnered with other audio developers, including DSP Concepts, best known for its embedded audio software platform Audio Weaver. Audio Weaver lets developers combine disparate audio processing software into one system, often with little or no coding needed. VoiceHub connects directly with Audio Weaver and works DSPC’s TalkTo software module for isolating human voices from other noise.
“Many of the brands building on Audio Weaver have been using Sensory’s TrulyHandsFree technology to power voice UI for years, so we anticipate significant interest in VoiceHub,” DSP Concepts CEO Chin Beckmann said. “Brands are looking for solutions that enable faster innovation, and bringing VoiceHub into Audio Weaver does just that. The user experience can be crafted and branded from day one, with accelerated time-to-market.”