While running pyannote/speaker-diarization-community-1, the embedding model's
minimum-input-size probe emits a PyTorch standard-deviation warning.
The captured stack identifies a synthetic random waveform, not the recording,
as the input responsible for this occurrence.