Voice setup
Get the voice subsystem dialed in.
ConstructOps runs voice on-device. This guide covers microphone permissions on iOS and Android, wake-word tuning for Spanish-first speakers, U.S. all-party-consent compliance, EN <-> ES code-switching, and voice biometric enrollment.
1. Microphone permission walkthrough
- On first launch, iOS prompts: "ConstructOps would like to access your microphone." Tap Allow.
- If you missed it, open Settings -> Privacy & Security -> Microphone -> find ConstructOps -> toggle on.
- Test the permission: open the app, tap the mic button, say a short phrase. The waveform should react.
- If the wake word does not work after granting permission, go to Settings -> ConstructOps -> Siri & Search -> Background App Refresh -> On. Apple requires background access for wake-word detection.
- Apple notes: voice recordings remain on-device. iOS shows a small orange dot in the status bar when the mic is active.
- On first launch, Android shows the permission card: "Allow ConstructOps to record audio?" Tap While using the app or Allow only this time.
- For wake-word detection, you need While using the app -- Only this time disables wake-word between sessions.
- If you missed it, open Settings -> Apps -> ConstructOps -> Permissions -> Microphone -> Allow only while in use.
- Test the permission: open the app, tap the mic button, say a short phrase. The waveform should react.
- Android 14+ shows a small green camera/mic indicator in the status bar when the mic is active.
2. Wake-word training tips for Spanish speakers
The default English wake word is "Hey Connixt." The Spanish variant is "Oye Connixt" -- both run on the same Picovoice Porcupine model. Speakers whose first language is Spanish often hit accuracy issues with the English "Hey" -- here is what helps.
- Train both wake words during enrollment. Settings -> Voice -> Wake word -> Add Spanish. Now both "Hey Connixt" and "Oye Connixt" trigger.
- Pronounce "Connixt" with stress on the first syllable: CON-nixt. Do not stretch the second syllable.
- For "Hey Connixt" -- the "H" sound is the most common stumble for Spanish-first speakers (Spanish "H" is silent). Practice 5 to 10 times during enrollment; the model adapts to your accent.
- Wake-word distance: 6 feet indoors, 4 feet outdoors with normal jobsite noise. If you have to lean close to the phone, retrain in the actual jobsite environment.
- If you switch languages during the day (start in English, switch to Spanish), the wake word follows. Just say "Oye Connixt" instead of "Hey Connixt" -- the rest of the command can be in either language.
- If wake-word accuracy drops over time (e.g., you got a new respirator that muffles your voice), retrain: Settings -> Voice -> Retrain. Takes 90 seconds.
3. Voice consent in all-party-consent states
Twelve U.S. states require all parties to consent before a conversation is recorded. ConstructOps voice runs locally and discards raw audio in under 60 seconds, but to be clear about your legal posture, the app shows a one-time consent screen the first time you record voice in any of these states.
4. Bilingual EN <-> ES code-switching
You can mix English and Spanish in the same sentence. The on-device speech recognizer handles code-switching natively. Common examples below -- all of these parse correctly.
5. Voice biometric enrollment (opt-in)
Voice biometrics let the kiosk recognize an individual worker from a 3-second voice sample at clock-in. This is opt-in -- workers must affirmatively enroll, and consent applies under Illinois BIPA (Biometric Information Privacy Act) and similar state laws (TX, WA).
- Open Settings -> Privacy -> Voice biometrics.
- Read the disclosure (English or Spanish). The disclosure lists: what is collected (a voice template, not raw audio), the purpose (identification at clock-in), the retention (deleted within 30 days of employment termination), and your right to withdraw at any time.
- Tap Agree and sign with finger or stylus.
- Speak the 5 enrollment phrases (about 90 seconds). The phrases are randomized -- two passes if your environment is noisy.
- The voice template is encrypted and stored on the kiosk device + Confidential Ledger. No raw audio is retained.
- You can withdraw at any time: same screen, Withdraw enrollment. The template is permanently deleted within 24 hours.