Introduction

This repository is a resource collection for Keyword Spotting (KWS), also known as Hotword or Wakeword Detection. KWS is used extensively in voice-activated personal assistants such as Alexa and Google Assistant. The Repo includes a list of research papers, models, implementations, algorithms, and tutorials related to KWS.

Table of Contents

1. Getting Started
2. Research Papers
3. Models
4. Implementations
5. Algorithms
6. Tutorials
7. Tools and Libraries
8. ToDo

Getting Started

To get familiar with KWS, there needs to be a basic understanding of digital signal processing and machine learning. Here are some resources

• 📖 Introduction to Digital Signal Processing
• 📖 Machine Learning Yearning by Andrew Ng

Research Papers

• EfficientWord-Net
  • 📑 Paper
  • 📦 Repository
• MatchboxNet
  • 📑 Paper
  • 📦 Repository
• Streaming KWS on Mobile Devices
  • 📑 Paper
  • 📦 Repository

Models

• WakeNet by ESP
  • 📘 Documentation
• Deep Residual Networks for KWS
  • 📘 Documentation
  • 📦 Repository

Implementations

• Keyword Transformer
  • 📦 Repository
• KWS Streaming
  • 📦 Repository
  • 📘 Inference on ESP32

Algorithms

• Beamforming
  • 📑 Paper
• Dynamic Time Warping
  • 📑 Paper

Tutorials

• Building Your Own KWS Model
• TensorFlow Lite

Tools and Libraries

• TensorFlow Lite for Microcontrollers
  • 📘 Documentation
• PocketSphinx
  • 📘 Documentation

ToDo

• Add accuracy and parameter diagrams for Broadcasted Residual Learning paper.
• Add performance table for PhonMatchNet paper.