voyelle-io/tinykw

TinyKW, short for Tiny Keyword, is a small engine for detecting specific spoken words, the kind of technology behind wake words like the phrase that turns on a voice assistant, or simple spoken commands. It is designed to be light enough to run on small embedded chips known as cortex-M microcontrollers, as well as on a Raspberry Pi or an ordinary desktop computer. Before you can use it, you first need to encode the words you want it to listen for using a web-based tool the project provides. That tool turns each keyword into a binary representation that gets pasted directly into your code. The project itself is written with a C interface at its core, with precompiled binary libraries and a C header file provided so you do not have to build it yourself. There is also a Python binding that mirrors the same C interface closely, for anyone who prefers working in Python. Using it in code follows a simple pattern: you initialize the engine, register one or more keywords along with a detection sensitivity threshold, then repeatedly feed it audio samples frame by frame. After each frame, you can check whether a specific keyword was detected, and clear that detection flag once you have handled it. Example code for both the C and Python versions is included in the project to show this pattern in practice. TinyKW is free to use for non-commercial purposes such as research, education, prototyping, and evaluation. Anyone wanting to use it commercially is asked to contact the project's author directly rather than assuming the free license covers that use.