Voice input¶
Talk to your robot. The agent transcribes locally and you tap send.
How to use¶
- Tap the microphone icon in the chat composer (between the
+and the send button). - The button enlarges and shows a live waveform. Speak.
- Release / tap again to stop.
- The transcript appears in the composer text box.
- Edit if needed, then tap Send.
Why tap-to-send instead of auto-send¶
The transcript step is a quality gate. SpeechRecognizer occasionally
mishears domain terms ("amcl" → "Amazon", "tf2" → "to" twice).
Letting you read and edit before send is one second of friction in
exchange for not chasing transcription artifacts in your conversation
history.
Permissions¶
First tap asks for the microphone runtime permission. Android shows the standard "Allow Scry to record audio?" dialog. Deny and the mic button stays inert with a toast on subsequent taps.
You can manage this anytime under Settings → Permissions in the app, or in Android Settings → Apps → Scry → Permissions.
Where the audio goes¶
Android SpeechRecognizer — Google's standard on-device API. On most
modern Pixels and Samsungs from 2022+ the recognition runs fully
on-device (Live Caption uses the same model). On older phones it may
fall back to a Google network call.
Scry doesn't ship audio anywhere. The agent only sees the final text transcript, just like if you'd typed it.
When voice fails¶
Mic button is greyed out / inert
Microphone permission denied. Go to Settings → Permissions → Microphone → tap to open Android's permission page → allow.
Transcript is wildly wrong
A few things to try:
- Speak in shorter chunks — long monologues compound errors
- Pause briefly between technical terms — "ros — topic — list" transcribes better than "rostopiclist"
- Type robot/topic names instead of saying them — recognizer models are weak on terms with slashes and underscores
No transcript appears at all
Google's SpeechRecognizer service may be disabled or uninstallable on some Android variants (Huawei devices without GMS, custom ROMs). Use the keyboard instead.