Wake-word confirmation

The voice-assistant example acknowledges a detected wake word with a 100 ms, 1 kHz beep. It retains microphone pre-roll, requests the tone, and starts the voice pipeline without waiting for playback to finish. Users can begin their question while the beep sounds: microphone capture and AEC continue, and speech captured during handoff stays queued until the listening stream is ready.

micro_wake_word:
  on_wake_word_detected:
    - lambda: id(voice_microphone).request_pre_roll();
    - lambda: id(voice_audio).play_wake_beep();
    - voice_assistant.start:
        wake_word: !lambda return wake_word;

Merge these actions into the existing wake-word handler; keep its UI and session setup actions. The tone request does not invoke TTS callbacks or response microphone muting and does not arm the barge-in gate.

Playback and reference

The playback task generates the tone with 5 ms fades, mixes it before playback gain and the reference tap, then sends it through normal speaker TX. The hardware reference captures the actual output. The cue bypasses TTS rebuffering and allocates no memory or waits in the wake callback. Clearing playback cancels it.

Cue level follows playback_gain_db and the DAC volume. The standard example keeps the DAC at the configured assistant volume while listening. Integrations that mute the DAC during listening can use id(voice_audio).is_wake_beep_active() to temporarily allow the configured volume, then restore their volume policy in a separate script without delaying pipeline start or microphone handoff.

Host tests verify duration, fades, mono/stereo output, saturation, cancellation and reference-tap placement. On hardware, check audibility and speech recognition when the user speaks during the beep. Host tests cannot establish acoustic cancellation quality.