Audible "POP" or *CRACK* after the final spoken sentence.

[voultar1]

I'm experiencing an audible "POP" or "CRACK" after the final sentence is spoken by the AI assistant in Home-Assistant.

The audible pop/crack doesn't happen at any other time, only when the final sentence is spoken, maybe 300ms after speaking is finished. I've tried changing voices, sampling rates, formats, etc.

[beecho01]

I looked into this but don't have a confident code fix to offer yet, so I'm leaving this open rather than guessing.

What I found:

• The integration itself does no audio post-processing — it forwards the raw bytes returned by the Kokoro FastAPI /v1/audio/speech endpoint straight to Home Assistant's TTS/media playback pipeline.
• There's no API parameter on the Kokoro FastAPI server to add start/end silence padding or control this — it's an open, unresolved feature request upstream (remsky/Kokoro-FastAPI#169).
• Since it reproduces on both the main Assist pipeline and an ESP32 satellite, and happens ~300ms after audio finishes (not during), this looks more like a playback-side artifact (e.g. an amplifier/DAC pop when the audio stream closes) than something encoded in the audio data itself — which would make it outside what this integration's code can control.

To help narrow it down further, could you share:

1. What media_player/speaker hardware plays the audio (brand/model, and whether it's the same for the ESP32 satellite)?
2. Your Home Assistant Core version.
3. Whether the pop happens with a very short test phrase too, or only after longer responses.

If it turns out to be format-specific (e.g. only on mp3/opus, not wav/pcm), that would point back at the integration/server rather than the speaker — worth confirming since you mentioned trying different formats already.