Mini toolsMOSS Transcribe Diarize
By OpenMOSS
Model family: MOSS
MOSS Transcribe Diarize is a unified audio text multimodal model built for Speaker Attributed, Time Stamped Transcription in a single end to end framework, jointly modeling lexical content, speaker attribution, and timestamp prediction. It is trained on extensive real world data and equipped with a 128k context window supporting inputs up to 90 minutes, allowing it to scale to long meetings, podcasts, and audiovisual content. Optional acoustic event annotation is also supported. It handles highly overlapping multi speaker dialogue, informal slang, regional accents, noisy environments, high dynamic emotional speech, and rapid speaker turn taking across Chinese, Japanese, and English content. Intended use cases include meeting minutes generation, call analytics, and long form audio and video content processing, providing a stable and reliable foundation for multi speaker transcription workflows.
Audio
Overview
MOSS Transcribe Diarize is an end to end multimodal large language model that jointly performs speech transcription, speaker diarization, and timestamp prediction. It projects multi speaker acoustic representations into the feature space of a pretrained text LLM, producing time aligned transcripts with speaker labels such as S01 and S02. It supports a 128k context window for inputs up to 90 minutes and handles overlapping multi speaker dialogue across multiple languages.
Last updated: July 9, 2026
