Robust Speaker Identification Model on CrossLink NX-33

Aizip’s efficient neural network based speaker identification model deployed onto the Lattice hardware.

Related Products

An efficient neural network-based speaker identification model was developed and deployed by Aizip Inc. The model can register multiple speakers and provide accurate and prompt identification results with low latency. It was co-designed to leverage the superior capabilities of the Lattice CrossLink NX-33 board, featuring an optimized hardware accelerator to support the model's computations.

Features

  • Smallest production ready speaker ID model
  • RISC-V and Hardware accelerator
  • Low false positive and false negative rate
  • Accurate, robust, and low latency model performance
  • Small memory footprint
  • Generic I2S audio interface
  • Optimized computation kernel

Block Diagram

Block Diagram

Board Photos

Board Image - Bottom

Board Image - Side

Board Image - Top

Kit Contents

  • HiLetgo SPH0645 I2S MEMS Microphone Breakout Module