Each sentence or 'segment' in whisper would be preceded by a [speaker] notation. This commit fixes that to only include the speaker in front of a larger group (since a new speaker would start a new diarization group this will always work).