Explore a tinyML Talks webcast on benchmarking and improving neural network execution on digital signal processors for hearing instruments. Learn about Demant's DSP-based platform and a custom neural network processing engine developed for power optimization. Discover performance comparisons between DSP and NNE implementations using a keyword spotting application benchmark. Gain insights into techniques like reduced word length, parallel MACs, and two-step scaling for enhancing NN execution in hearing aids.
Benchmarking and Improving NN Execution on Digital Signal Processors for Hearing Instruments
Overview
Syllabus
Introduction
Sponsors
Upcoming talks
First talk
About Zuzana
Overview
Team
Diamond
Presentation Overview
Motivation
Deep Neural Network
Generic Audio DSP
Reduced Word Length
Multiply Accumulate
Memory
Scaling
Results
Conclusion
Questions
Thanks Sponsors
Taught by
tinyML
