What can the wake phrase do by itself?
Porcupine listens for a phrase and tells your app when it finds one. Your app decides what happens next, such as showing a button to start dictation. The wake detector does not turn the rest of the speech into words. Give people a way to use the feature without the microphone.
What runs on the device, and what still needs an account?
The wake detector processes audio on the device. Setup requires a Picovoice account and access key. Choose the software package and phrase file for the device you support. Picovoice says it collects usage metadata. Check the connection requirements for your chosen package and track separately whether any later speech step sends audio elsewhere.
How would you estimate use before shipping?
For Porcupine, the vendor counts monthly active instances of the engine. That can mean a device, app or browser instance rather than a person with an account. The free trial is one-time. Check the device count and terms for the app you intend to distribute before relying on that allowance.
What do you get back from detection?
The detector tells your app which phrase matched. The optional custom phrase setup produces a model file for the target platform. Keep the supported model file and the phrase settings with the project, and decide separately whether the app needs to save any later audio or text.
What should happen when a similar phrase is heard?
Sketch this with fictional material. Leave the real site, accounts and services alone.
- Draw the idle screenMark Start voice and a separate button for entering text. Decide how a person can tell whether the microphone is active.
- Compare two phrasesWrite the intended trigger and a similar phrase. Give the expected result for each and say what the user can do if the app gets it wrong.
- Mark the data boundaryDraw the trigger notification leaving the detector. Add a separate arrow only if a later step needs audio or a transcript.