Skip to main content

Hold-to-Talk Mode

Hold-to-talk is the chat mode where you press and hold the button to record, then release to stop and upload. The capture window lasts exactly as long as you hold the button — nothing starts or stops on its own.

It is one of the four voice chat modes; register it with ai_mode_hold_register().

When to use it​

Use hold-to-talk when you want deliberate, explicit turns and full control over what the device hears:

  • Noisy rooms — the microphone only captures while the button is down, so background speech and noise outside that window never reach the cloud.
  • Deliberate turns — the user decides exactly when a turn begins and ends, with no wake word and no voice-activity guesswork.
  • No false triggers — nothing is uploaded until a button is pressed, so the device never reacts to ambient sound.

The trade-off is that it is hands-on: every turn needs a physical press. For hands-free interaction, use wake-word or free mode instead.

How it behaves​

A turn follows the shared mode lifecycle. Pressing the button moves the device from IDLE into LISTEN; releasing the button ends capture and advances it through UPLOAD, THINK, and SPEAK before it returns to IDLE.

note

Capture is bounded by how long you hold the button. The mode needs the button component (ENABLE_BUTTON) to receive press and release events.

Enable it​

Register the mode at startup, then make it the active mode with ai_mode_init:

ai_mode_hold_register();
ai_mode_init(AI_CHAT_MODE_HOLD);   // AI_CHAT_MODE_HOLD | ONE_SHOT | WAKEUP | FREE

See Voice Chat Modes for the full startup sequence — registering several modes, running the task loop, and switching between them at runtime.

See also​