porcupine

Why we included this project

The hard part of a voice-enabled product is often the wake word, not the speech recognition that follows it: the listener has to stay on constantly without draining the battery or hogging the CPU. Porcupine handles that with a deep-learning engine that detects a chosen trigger phrase entirely on-device, so there is no round-trip to a server before the rest of your app wakes up. It is light enough for small microcontrollers yet also runs on phones and in browsers, and official bindings for Python, Java, Swift, JavaScript, and other languages let a team reuse the same detection instead of hand-rolling a hotword solution. Custom wake words can be trained through Picovoice Console, which matters when the trigger has to be a brand name rather than a generic word. If low latency and a small footprint are what you are optimizing for, this is a solid, practical option.