Enhancing voice recognition for the automated assistance markets

01 June 2017

Credit: Zapp2Photo/Shutterstock

A semiconductor solutions provider have announced their development kit, which delivers enhanced audio processing to improve voice recognition rates.

This will enable the product’s (the AcuEdge Development Kit by Microsemi Corporation) use in adverse audio environments for emerging human to machine (H2M) applications in the internet of things and automated assistance markets. The development kit is designed for the Amazon Alexa Voice Service (AVS) to enable third party developers and ODMs to evaluate and incorporate Alexa functionality in H2M applications – while interfacing with a multi-mic processor, also by Microsemi.

Microsemi’s Timberwolf is a field-programmable, field-upgradable digital signal processor equipped with specific hardware accelerators to run three simultaneous acoustic echo cancellers (AECs). The AcuEdge Development Kit comes complete with a development board that connects directly to a Raspberry Pi and plastic frames to help position microphones and speakers in targeted applications.

Microsemi’s multi-mic processor is enabled with compressor/limiter/expander microphone pickup algorithms, beam forming, and stationary and non-stationary noise reduction, enabling voice control and speech recognition (from a distance and also in the presence of adverse noise and audio sounds). The complete kit is populated with two microphones which can operate in both 180 degree and 360 degree audio pickup modes. The evaluation board is equipped with the Raspberry Pi connector, which allows for quick and easy setup. Additionally the kit is delivered with a board and speaker stand allowing users to quickly evaluate the performance of the solution.

“Amazon’s goal is for developers and device makers to have access to a wide range of reference solutions that fit their needs,” said Priya Abani, Director, Alexa Voice Service. “We’re excited to be working with Microsemi to provide device manufacturers with a way to quickly and easily integrate Alexa into their products across a variety of form factors, including wall-mounted and on-table devices, smart speakers, and connected home products.”

While the Raspberry Pi runs the Amazon AVS client and Alexa wake word detection, the Timberwolf solution runs all front-end audio processing, enabling significantly improved speech recognition accuracy.


Print this page | E-mail this page