Key Phrase Detection

Lattice sensAI Demo

Our system is going under maintenance starting April 18, 2025 at 6:00 PM Pacific and ending April 20, 2025 at 4:00 PM Pacific. During this window, the website may not be reachable. For immediate assistance, please contact techsupport@latticesemi.com.

Convert voice commands into system actions – This demo provides an example for key phrase detection on iCE40 UltraPlus Ultra-low power FPGA. A Convolutional Neural Network (CNN) acceleration engine is trained to with thousands of voice samples to detect one of four key phrases.

On device Always-on listening – Connect a digital microphone directly to the Lattice inference engine and enable always-on listening with key phrase detection, no need for cloud connectivity to detect key phrases.

Neural Network on a 5.5mm2 FPGA – The Lattice inferencing engine fits into two package options in our iCE40 UltraPlus FPGA. A 30-ball CSP package with 0.4 mm ball pitch created the smallest neural network within an FPGA, 2.15 mm x 2.55 mm. A 48-pin QFN package with 0.5 mm pin pitch enables lower cost PCB designs, 7.0 mm x 7.0 mm.

Features

  • Able to detect one of four key phrases using the iCE40 UltraPlus HM01B0 UPduino Shield
  • 6 layer CNN trained with thousands of sample key phrases
  • Runs at 40 evaluations per second consuming 7 mW
Lattice sensAI

Jump to

Block Diagram

Key Phrase Detection Block Diagram

Documentation

Quick Reference
Downloads
TITLE NUMBER VERSION DATE FORMAT SIZE
Select All
Key Phrase on HM01B0 UPduino Shield Demonstration User Guide
FPGA-UG-02094 1.0 10/24/2019 PDF 944.8 KB
MDP Based Key Phrase Detection Demonstration User Guide
FPGA-UG-02048 1.1 9/25/2018 PDF 1.1 MB
TITLE NUMBER VERSION DATE FORMAT SIZE
Select All
MDP Based Key Phrase Detection Demonstration Bitstreams
1.1 9/25/2018 ZIP 2.1 MB
Key Phrase on HM01B0 UPduino Shield Demonstration Bitstreams
1.0 10/23/2019 ZIP 142.5 KB

*By clicking on the "Notify Me of Changes" button, you agree to receive notifications on changes to the document(s) you selected.

Like most websites, we use cookies and similar technologies to enhance your user experience. We also allow third parties to place cookies on our website. By continuing to use this website you consent to the use of cookies as described in our Cookie Policy.