Speech Recognition on Low Power Devices

Explore speech recognition technology for low-power devices in this tinyML Talks webcast featuring Fluent.ai experts Vikrant Tomar and Sam Myer. Delve into the process of transitioning from high-level libraries like Pytorch to running models on ARM Cortex M series microcontrollers and DSPG digital signal processors. Discover optimization techniques including low-level programming, 8-bit quantization, unique model architectures, network compression, and layer selection. Learn about multilingual speech recognition, automatic intent recognition, and system challenges. Witness live demonstrations of smart home applications and compare streaming microcore performance with TensorFlow Lite. Gain insights into memory requirements and strategies for running larger models on smaller devices in this comprehensive exploration of speech recognition technology for resource-constrained environments.

Introduction
About tinyML
Multilingual speech recognition
Automatic Intent Recognition
Demo
Sam
System Overview
Common Challenges
Model Compression
DTPG
Wakeboard
Microcontrol library
Streaming
Streaming example
Memory requirements
Streaming Microcore vs Tensorflow Lite
Running larger models on smaller devices
Live demo
Smart home demo
Questions
Thanks