I Turned a Product Name into an Offline Wake Word — What I Had to Test
Most smart products already have a name. The hard part is making the device hear that name reliably without sending every conversation to the cloud. This is the product question behind a custom wake word: can a device recognize one short phrase locally, then start the next interaction? The product name and the wake word are different things A brand name is designed for people. A wake word is…
Turning a product name into an offline wake word involves testing the phrase's reliability under various conditions. The product name and wake word are distinct, with a brand name designed for people and a wake word aimed at a detector. For instance, a robot named "Taro" would require the phrase "Hey Taro" to initiate actions. Keyword spotting (KWS) is a suitable option if the product only needs to recognize one wake phrase or a few commands, rather than general speech understanding.
Before choosing a wake word, testing with multiple speakers, varying speeds, distances, and background noise is crucial. The process begins by selecting a product name, training a custom keyword model, testing it with real voices, and then exporting it in a format like ONNX or TFLite for local execution on target devices. This model should be compatible across Python, Web, Android, and embedded hardware.
The project documentation should include the model, reproducible inference example, supported sample rate and audio format, default threshold and cooldown settings, and a test checklist for the final microphone and enclosure. It's important to note that online testing is merely a screening step, and the final product will still need to be measured for recall, false triggers per hour, latency, memory, and power consumption.
The ONNX or TFLite model, along with the necessary runtime and examples, is available in onnx-wakeword.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
