English speech as Mimi audio tokens with separated speaker tracks, transcripts with word timings and emotion labels.
Note Pre-training: 408 hours from YODAS3.