DeepGate is a speech recognition system that brings together machine learning and re-programmable logic devices. By implementing a pre-trained neural network using a Spartan-6 FPGA, we are able to classify a limited vocabulary with nearly perfect accuracy. Our system is capable of recognizing the numbers zero through nine as well as the cardinal directions. In addition, we designed a custom PCB that housed the FPGA and allowed it to communicate with a Raspberry Pi. The Pi is responsible for providing a GUI and demo platform.
We would like to thank SoarTech and Joshua Haley (joshua.haley@soartech.com) for sponsoring our project.