decktalk narrate reports it when it makes or keeps the narration takes.
Pass --allow TAKE_WORD_CUT to carry on past it, and --fail-on to choose which severity fails a run.
The settings that move this verdict are narration.sound_end_noise_dbfs, narration.sound_end_min_run_seconds.
What to do
Lowernarration.sound_end_noise_dbfs so the take’s sound end is measured after its last word, then run decktalk narrate again.
Troubleshooting says more under A take that plays silence or cuts a word.