Mac App Issues — Talk Mode Audio + Voice Wake (v2026.3.12, macOS Tahoe 26.3, arm64)

[sullen fiber]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[keen current]

A couple things to sanity-check here, because Talk Mode TTS playback and “normal” /tts audio file generation are actually different paths.

1) Talk Mode not playing audio (but files are fine)

Talk Mode on macOS uses ElevenLabs streaming + incremental playback (not “generate a file then play it”), and it’s driven by the talk settings.

Things that commonly make it look like “no playback”:

• interruptOnSpeech instantly stops playback (mic picks up your own speaker / room noise and immediately interrupts).

  • Quick test: in the Mac app Talk Mode settings, temporarily set Interrupt on speech = off and try again (or use headphones).

• Output format mismatch / MP3 streaming edge case

  • On macOS the documented default is PCM (pcm_44100). If you’ve forced MP3 (mp3_*) in config, try removing it or explicitly setting:

    {
      talk: {
        modelId: "eleven_v3",
        outputFormat: "pcm_44100"

}
}
```
Docs: https://docs.openclaw.ai/nodes/talk

Also: are you running the Mac app in Local (this Mac) vs Remote over SSH / Direct? (Because “restart the gateway” won’t necessarily touch the Mac-side audio pipeline.) Docs: https://docs.openclaw.ai/platforms/mac/remote

2) Voice Wake stops listening after the first successful send

Per the macOS behavior, after a send the recognizer should restart automatically. If it doesn’t, one big clue is:

• Does the overlay ever get “stuck” visible after the first send?
  • If yes: hit the X to dismiss it and see if listening resumes. There used to be a “sticky overlay blocks restart” failure mode (should be fixed, but could be regressing on your build).
    Docs: https://docs.openclaw.ai/platforms/mac/voicewake

3) The fastest way to get us to root cause: grab the Mac app diagnostics log

The Mac app can write a rolling JSONL log: