README for Keyword Spotting Project

Introduction

The Keyword Spotting project is a real-time system for recognizing specific spoken words or phrases. Utilizing deep learning, it analyzes audio inputs to identify and classify spoken keywords, suitable for applications such as voice-activated assistants and home automation systems.

Quick Start

Hardware Requirements

• Sony Spresense main board: The core microcontroller unit for processing.
• Spresense extension board: For additional connectivity and features.
• Microphone: Following the tutorial for using multiple microphone inputs with Spresense.
• OLED Display (SSD1306): For real-time display of keyword spotting results.
• 32GB Micro SD Card: Used to store the DSP driver for audio processing.

Software Dependencies

While environment_install.sh takes care of setting up the software environment, it is important to be aware of the key dependencies:

• Python 3.8 or later
• PyTorch
• ONNX
• NumPy
• Additional Python libraries listed in requirements.txt

How to Install Dependencies

Install the required Python libraries by executing the following command in your terminal:

bash ./environment_install.sh

Preparing the Dataset

Download the dataset from this link and unzip it in the project's root folder. Ensure that the data directory structure is correct as expected by the scripts in the project.

How to Run the Project

Execute the run.sh script to run the entire project, which includes steps for training, testing, and model conversion:

bash ./run.sh

How to Deploy on Arduino

The deployment on Arduino is facilitated through the run.sh script, which prepares the model in a compatible format:

0. Follow this intruction to set up your arduino Board Manager: https://github.com/zhenyulincs/Sony-Spresense-Arduino-TFMicro
1. Run the run.sh script to ensure the model is converted to a TensorFlow Lite format and then to a C header file.
2. Integrate the generated C header file with your Arduino sketch and upload it to the Arduino environment.