MOSS-Transcribe-Diarize

★1,384

Tiny open model that does long multi-speaker transcripts in one pass.

Why it matters

Meeting and podcast pipelines still glue separate ASR and diarization stacks together. MOSS-Transcribe-Diarize is a 0.9B open model that jointly outputs speaker labels, timestamps, and text for long messy audio, and it just took first place in the INTERSPEECH 2026 MLC-SLM challenge across 14 languages. The practical win is one end-to-end open path for speaker-aware transcripts without a fragile multi-stage pipeline.

speechopen-weightsmultimodal

Source ↗

Via github