This repository contains the implementation and deployment codes for my EIE4433 Honours Project at The Hong Kong Polytechnic University (PolyU). The project focuses on deploying a lightweight Deep Learning model for real-time speech recognition on embedded edge devices.
Important
This project was selected as one of the "2023 Outstanding Works by Students" by The Hong Kong Polytechnic University.
The work is officially showcased in the PolyU Library collection: 🔗 View Project Showcase: Smart Security Robot
This project demonstrates the end-to-end pipeline of training a speech recognition model and deploying it on a Raspberry Pi using TensorFlow Lite (TFLite).
Key Features:
- Model Architecture: Lightweight CNN/RNN optimized for edge inference.
- Optimization: Quantization (Float16 / Int8) to reduce model size and latency.
- Hardware: Raspberry Pi (3B+/4B) with USB Microphone.
- Performance: Real-time inference with low latency (<100ms) and high accuracy on keyword spotting tasks.
- Raspberry Pi (3B+ or 4B recommended)
- USB Microphone (or ReSpeaker HAT)
- MicroSD Card (16GB+ with Raspberry Pi OS)
- Python 3.7+
- TensorFlow Lite Runtime
- PyAudio / SoundDevice
- NumPy
Open the terminal on your Raspberry Pi and clone this project:
git clone [https://github.com/Zilai-WANG/EIE4433-Speech-Recognition-of-tensorflow-Lite-on-Raspberry-pi.git](https://github.com/Zilai-WANG/EIE4433-Speech-Recognition-of-tensorflow-Lite-on-Raspberry-pi.git)
cd EIE4433-Speech-Recognition-of-tensorflow-Lite-on-Raspberry-pi