Key Phrase Detection

Lattice sensAI Reference Design

Customizable to detect any phrase - This reference design continuously searches for a key phrase utterance via a digital MEMS microphone. By updating the training dataset using deep learning frameworks such as Caffe, Tensorflow or Keras, designers can add wake word capability to a system

Optimized Power and Performance - The reference design enables a flexible system design trading off speed and power, the system can be put in low power listening for voice mode consumer as low as 0.6 mW.

Complete Reference Design - Training dataset, scripts for training using common Neural Network training tools, and NN model are provided to enable modification.

Features

  • Accelerated, low-power key phrase detection at the network edge using neural network model
  • iCE40 UltraPlus
    • VGG8 like 8-bit CNN
    • Sound samples converted to histogram and inputted to Neural Network as image
    • Up to 40 evaluations per second
    • 7 mW of power consumption
  • Adjustable frame rate. Can be optimized between power and response time depending on system needs
Lattice sensAI

Jump to

Block Diagram

Documentation

Quick Reference
Technical Resources
TITLE NUMBER VERSION DATE FORMAT SIZE
iCE40 UltraPlus Key Phrase Detection Quick Start Guide
FPGA-AN-02008 1.0 10/24/2019 PDF 1.4 MB
TITLE NUMBER VERSION DATE FORMAT SIZE
Key Phrase Detection Using Compact CNN Accelerator IP - Project Files
1.0 10/23/2019 ZIP 1.2 MB
Key Phrase Detection Using Compact CNN Accelerator IP - Documentation
FPGA-RD-02066 1.0 10/24/2019 PDF 2.7 MB


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.