This repository provides a comprehensive framework for implementing keyword spotting on STM32 microcontrollers. The key components of this framework include:
- PyTorch Framework: Utilizes a filter bank to extract features from audio data and trains an RNN-GRU network.
- TensorFlow Lite Framework: Facilitates the training of a Depthwise Separable Convolutional Neural Network (DS-CNN). It supports Quantization-Aware Training (QAT) to optimize the network and generates a C header file for deployment on the MCU.
- STM32 Project: Contains the setup for deploying and running real-time keyword spotting on STM32 hardware. (Not public)
- Demo Scripts: Provided to facilitate running the demonstrations directly on the MCU.
For detailed instructions, please refer to the README.md files located in the respective directories within this repository.