tinyML Asia 2021 Shaorui Huang (mandarin & english): Powering innovation of low-power smart…
tinyML Asia 2021
Powering innovation of low-power smart speaker with Voice-Dedicated AI Chip
Shaorui HUANG, General Manager of Product R&D Center, Allwinner Technolgy
More and more application scenarios are migrating AI computing from the cloud to embedded devices. Due to the diversification of application requirements, the main problem embedded artificial intelligence hardware is facing is how to balance power、performance and cost. In applications, we also need to achieve a unified deployment and operating environment based on different deep learning frameworks. In the past few years, we have been focusing on the research of smart speaker chip solutions. In order to deal with the challenges of the continuous increase in power consumption and performance of smart speaker products, we have developed a dedicated chip for smart speaker products, using a heterogeneous architecture of CPU+DSP+AIPU, based on deep learning frameworks, which can maximize the performance of parallel computing for voice AI，and provide better performance at lower cost and power consumption.
source
