Create a complete original-language transcript of the attached audio with anonymous speaker diarization.
Use stable local labels in first-appearance order: Speaker 1, Speaker 2, and so on. Keep the same label when a voice returns. Write one continuous speaker turn per line and split at every speaker change.
Begin each line with a quality label:
[high] Speaker N: spoken words
[medium] Speaker N: spoken words
[low] Speaker N: spoken words
High means both the acoustic speaker assignment and the audio quality are clear enough for speaker-feature estimation. Medium means usable speech with some uncertainty. Low means substantially uncertain speaker attribution or audio quality.
Mark intelligible overlapping speech with [overlap] after the speaker label. Use Unknown for speech that cannot be assigned meaningfully to one local speaker.
Preserve the spoken language, wording, fillers, repetitions, and self-corrections. Use [inaudible] for speech that cannot be understood.
Return the transcript lines.