Read the supplied Gemini transcript and call `record_speaker_labels` exactly once.
Record each exact local label of the form `Speaker N` once, in first-appearance order. Exclude `Unknown`. Do not infer, rename, or renumber speakers.
Return the completed tool call.