OpenMOSS/MOSS-Transcribe-Diarize

A 0.9B model for long-form transcription in 50+ languages with speaker diarization, timestamps, and acoustic event awareness

★ 2,099Forks 131PythonGitHub ↗Compare

Project website ↗

audio automatic-speech-recognition multilingual speaker-diarization speech-recognition subtitles transcription

Issues

Transcript collapses to a single marker-less segment when the audio ends with ~0.25s+ of near-silence (deterministic, content-dependent)

#61 · open · 0 comments

Diarization: sub-second utterances produce noise embeddings (one speaker became four identities)

#60 · open · 0 comments

Model returns a full refusal sentence when the window is digital silence

#59 · open · 0 comments

超出40分钟的音频会提前结束转录

#57 · open · 3 comments

CUDA out of memory.

#55 · open · 2 comments

解析一段60分钟的电话录音，16G的显存够用吗？

#53 · open · 3 comments

Repo id must use alphanumeric chars, '-', '_' or '.'. The name cannot start or end with '-' or '.' and the maximum length is 96: 'OpenMOSS-Team\MOSS-Transcribe-Diarize'.

#52 · closed · 0 comments

Timestamps inacurate

#51 · open · 2 comments

Word timestamp

#50 · closed · 1 comments

webUI is not functioning

#47 · open · 0 comments

CUDA out of memory error with long audio

#46 · closed · 1 comments

What should be the prompt to transcribe in another language than Chinese or English?

#45 · open · 0 comments