MOSS-Transcribe-Diarize 0.9B

A 0.9B end-to-end audio-to-structured-transcript model from OpenMOSS that produces speaker-attributed, time-stamped transcription (SATS) in a single generation pass — no separate ASR + diarization pipeline. Handles up to ~90 minutes of audio in a 128k context, labels speakers anonymously ([S01], [S02]), emits start/end timestamps, is acoustic-event aware, and supports hotword biasing for names and domain vocabulary across 50+ languages. Won 1st place in the 2nd MLC-SLM Challenge at INTERSPEECH 2026 across 14 languages.