MOSS Transcribe Diarize
OpenMOSS · released May 19, 2026 · OpenMOSS-Team/MOSS-Transcribe-Diarize
Speech to textTranscribes a recording into words
- Type
- Open weightsApache License 2.0
- Languages
- 2
- Size
- 0.9B
Context measured in tokens
Our take
Written Sep 29, 2026MOSS Transcribe Diarize is a small speech-to-text model you can download and run yourself, covering English and Chinese. It is strongest on the messy recordings — meetings, podcasts, accented speech — and weakest on clean, carefully read audio, where it sits near the bottom of the field.
Reach for it when your recordings are far-field and crosstalk-heavy, or when you are working through a long backlog: an hour of audio takes about 9 seconds on the leaderboard's own hardware, though your machine may differ. Skip it if you need accurate transcription of clean, carefully read-aloud audio, if you need speaker labelling, timestamps or streaming, or if you wanted a hosted offer to try first.
The case for it
- Meeting recordings come out at 8.3% of words wrong, better than most of 92 models, so far-field audio with people talking over each other is where it holds up.
- Accented speech lands at 6.1% of words wrong, better than most of 76 models, which makes it a candidate for non-native speakers rather than a last resort.
- 381 times real time on the leaderboard's own hardware, an hour of audio in about 9 seconds, so a large archive can be worked through in one sitting.
- The licence allows commercial use, changes and redistribution (Apache License 2.0), so it can go into a product.
The case against it
- Clean read-aloud recordings come out at 1.7% of words wrong, worse than most of 92 models and 60th of 92 on Clean read speech as of 28 Sep 2026 — the easy case is where it falls down.
- Every accuracy figure we hold is English, so its Chinese coverage is unverified for accuracy.
- We list no host for it, so running it yourself is the only route we can point you to.
How good is it?
An open transcription model for turning accented speech and podcast or video audio into written text.
- transcribing speakers with a range of accentsAccented speech · 18th of 76
- transcribing podcasts and video audioPodcasts and video · 23rd of 92
TranscriptionTurning speech into text4 of 5Open ASR WER · 21st of 76
95.4%
Misses roughly one word in 22, averaged over nine English test sets.
381×53rd of 74
an hour of audio in 9 seconds, on the board's own hardware. Your machine will differ.
2
Listed on the model card. The accuracy above is English only.
Percentage of words wrong on each set, lower better. Bars are scaled to this model's own worst case; the placing beneath each rate is against every model measured on that set.
Which languages ↓ ↑
English · Chinese
The figures above come from the Open ASR Leaderboard, an independent public test that runs every model on the same recordings. It is the only measurement of transcription quality we know of, so there are no other scores to show.
Each of these is the same transcription job on a different kind of recording, so together they say where it holds up and where it slips — not how closely it follows an instruction.
Every published score for this model9 scoresEvery figure we hold, from 9 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
Comfortable fit
GeForce RTX 4090 · 24 GB
Room to spare. 20.8 GB spare means a 10% error in the size would not change the answer.
Apple M1 Pro (16-core GPU) · 32 GB
Room to spare. 22 GB spare means a 10% error in the size would not change the answer.
Apple M2 (8-core GPU, 8GB unified) · 8 GB
Room to spare. 4 GB spare means a 10% error in the size would not change the answer.
These cards answer whether MOSS Transcribe Diarize loads, not how fast it transcribes. Throughput figures for a transcription model come from its text decoder, so treat this as a fit answer rather than a speed one.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first. This is a fit answer: whether it loads, not how fast it transcribes.
When we formed this view
Recent changes
What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
Licence and identifiers
What the licence allowsApache License 2.0, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Apache License 2.0
Fully permissive: commercial use, redistribution, and derivatives allowed. Requires attribution and a copy of the license. Includes an express patent grant.
Identifiers
- Hugging Face
- OpenMOSS-Team/MOSS-Transcribe-Diarize
- Architecture
- Dense
- Takes in, gives back
- Audio in, text out
- Catalogue slug
- openmoss-team-moss-transcribe-diarize