Sensory and STMicroelectronics (STM) are trying to make it easier for developers to build products with embedded speech recognition capabilities. To that end, the two companies have a unveiled a new solution that combines the STM32 microcontroller (MCU) with Sensory’s portfolio of voice recognition products.
The two long-time partners believe that their joint solution will enable voice applications in a range of IoT devices, with a particular emphasis on wearable and smart home offerings. In that regard, they emphasized the fact that the STM32 has enough memory to store voice applications and speech models directly on the chip itself, which speeds up processing times and makes it easier to integrate the solution into the design of a new device. It also minimizes the amount of data that needs to get transferred, which reduces costs for developers.
Sensory, meanwhile, is an established provider of voice development tools. The company’s VoiceHub online portal allows clients to create application-specific speech recognition models with their own custom wake words and command sets, with support for nearly 20 different languages and dialects.
Sensory and STM are hoping that their solution will allow developers to prototype ideas and bring new products to market more quickly. The solution specifically leverages the STM32H7 MCU and the STM32Cube software extension package.
“This collaboration sets to jump-start the development of embedded-voice user interfaces,” said STM EVP and General-Purpose MCU Sub-Group VP Ricardo De Sa Earp. “The combination of ST and Sensory technologies will enable the STM32 user community to deploy ‘Voice AI on the edge’ without any programming, data-science, or machine-learning expertise, for free in prototypes and with favorable licensing terms in production.”
Sensory’s VoiceHub is built with its TrulyNatural voice platform, which is designed for on-device speech recognition. The company has also released the SensoryCloud AI-as-a-Service solution for clients that are looking for flexibility rather than edge processing.
June 10, 2022 – by Eric Weiss