Keyword spotting (KWS)

Detect wake words and voice command triggers locally. Cheaper than full speech-to-text when you only need to catch specific phrases. No full transcription, no cloud API, no API key.

KWS engines listen for a pre-defined set of keywords or phrases and fire when one is matched. Think “Hey Decibri” as a wake word, or a fixed command vocabulary like “stop, next, pause, play”. The engine discards everything else; it does not transcribe free-form speech.

When to use KWS vs STT

• KWS: you have a small, known vocabulary (wake word, menu commands, < 100 phrases). Lower CPU, lower latency, always on-device.
• STT: you need free-form transcription, dictation, conversational capture, or anything outside a fixed phrase list. See Speech-to-text integrations.

Learn more