Aizip Announces Innovative Advanced Neural Architectures for Audio Applications

Saratoga, CA, September 8, 2021 – Aizip, Inc., a Silicon Valley startup providing best-in-class tiny artificial intelligence (AI) models for the Internet of Things (IoT), or the so-called artificial intelligence of things (AIoT), today announced breakthrough deep neural network (DNN) architectures for diversified audio applications.

Audio is one of the first and most important areas for AI applications in the IoT market. Various types of AI models, such as keyword spotting (KWS), noise reduction, and speaker identification, can be found in intelligent homes, smart cities, and industrial automation, among other applications. The AI community at large is expending significant effort to migrate these models from cloud to edge devices to provide low-cost solutions, instantaneous response, and data privacy. Despite much recent progress, most audio AI models still require sizable IC chips for their processing power and memory. However, wide adoption demands much smaller neural network models on small IC chips to achieve low cost and low-power operation.

Aizip has now announced the availability of new DNN architectures to cover a wide range of audio applications. One of the innovative networks is Zenet for noise reduction. Another novel network architecture, Barrelnet, is for speaker identification. These tiny machine learning (TinyML) models, together with other audio inventions, have been designed with a focus on environments that have much less processing power and memory. In some application scenarios, these models can achieve a 100x reduction in power and memory while maintaining competitive performance.

“This is a major milestone towards ubiquitous usage of audio intelligence in the real world,” commented Boltzmann Li, who led the algorithm team at Aizip to bring audio TinyML models to low-cost ICs such as the ARM M series. Now the integrated hardware and software modules can be less than $1, with some of the modules well below $1, according to Li. He further shared a case where KWS has been successfully deployed on a Cortex M0+ chip and delivered to one of Aizip’s customers.

These breakthrough architectures will allow Aizip to expand its Aizip Intelligent Audio (AIA) portfolio with a suite of models covering a wide range of audio applications with superior specifications. The company, which is celebrating its first anniversary this month, says the demand for audio TinyML solutions is very strong and it is expanding to meet customer needs.

In addition to AIA, Aizip is also working to grow its TinyML models in Aizip Intelligent Vision (AIV) and Aizip Intelligent Time-Series (AIT). Aizip is committed to bringing low-cost and high-performance intelligence for the benefit of society worldwide. For more information, please contact info@aizip.ai.

* * *

About Aizip, Inc.

Aizip, Inc., develops AI models for edge applications. Based in Silicon Valley, Aizip provides models with superior performance, quick turnaround time, and excellent ROI. These models can be used in a wide range of applications for an intelligent, automated, and connected world. For more information, visit www.aizip.ai.