Sensory brings ChatGPT 3.5 to voice UI customization

A variety of companies are scrambling to investigate if and how they should integrate generative AI into their applications. For some, it is a stretch, but for others it is a slam dunk, and you can count Sensory, Inc. in the latter camp.

The provider of AI-based speech recognition technologies has integrated the ChatGPT 3.5 large language model with its own technology for creating voice-controlled user interfaces (UI) and assistants to power the latest version of its VoiceHub platform. VoiceHub 2.0 leverages generative AI capabilities from ChatGPT to strengthen the capabilities of its web portal for creating customized voice UIs, and radically streamline the amount of time it takes developers to prototype and produce these UIs.

“VoiceHub 2.0 will absolutely boost efficiency by dramatically reducing build time required to develop a full-blown voice assistant engine,” said Jeff Rogers, vice president of sales at Sensory. “Not only will VoiceHub 2.0 help our customers prototype new projects in a matter of minutes, it will help them cut big chunks of time out of their go-to-market strategy for new products.”

"Speech recognition is important in many applications, but not all needs are the same. Our scalable VoiceHub voice UI development tools were created to provide developers a one-stop-shop for applications ranging from ultra-small footprint embedded wake words and command models to full-featured NLU voice UIs,” said Todd Mozer, CEO of Sensory. “With VoiceHub 2.0, we’re providing an even more powerful, flexible, and intuitive tool that harnesses the power of generative AI to make short work of creating high-performance speech recognition models."

Specifically, the ChatGPT integration enables a new “Task Explorer” feature in VoiceHub 2.0 that streamlines the creation of large vocabulary or natural language (those based on the company’s TrulyNatural 6.21.0 technology) voice UIs. Leveraging generative AI allows the Task Explorer function to provide a variety of relevant intents, commands, or other options when a user enters the domain type of the project. Users would simply select all options that would be relevant to their product’s/project’s capabilities and features. Based on domain or product category, VoiceHub’s new Task Explorer feature can also generate a list of suggested phrases for the language model, which users can pick from to expedite voice user interface development. 

Among other updates in VoiceHub 2.0, the portal user experience has been refreshed with a new layout and new features that allow for sharing and importing of projects in an easy-to-use drag and drop format. The company’s TrulyHandsfree Micro 7.1.0 has been updated with speed improvements, and support for more hardware platforms, including ARM Cortex-M4, Silicon Labs Cortex-M33/M, Ambiq Apollo 4, Cadence Hifi5, Qualcomm, and Xmos xcore.ai. The technology also supports Android, iOS, Linux and Windows operating systems, and Sensory said support for other operating systems can be added upon request.

The company’s technology updates also extend to its TrulyNatural platform, with the 6.21.0 version featuring a background model for US English that improves out-of-vocabulary rejection, making it even easier for grammars created in VoiceHub to perform with high accuracy in real world applications. The TrulyNatural SDK 6.21.0 now requires less RAM for recognizers on small, embedded platforms and adds Voice Activity Detectors for the SNSR-lite Large Vocabulary Continuous Speech Recognizer. That all means VoiceHub-generated grammars using this new version can be used on a wider variety of devices.

Sensory also said VoiceHub 2.0 now supports 25 languages and regional dialects.