Models / ESPnet/ Owsm CTC v3.2 ft 1B

Owsm CTC v3.2 ft 1B

ESPnet · released Sep 24, 2024 · espnet/owsm_ctc_v3.2_ft_1B

Speech to textTranscribes a recording into words

Input: audio. Output: text.InputOutput
Type
Open weightsCreative Commons Attribution 4.0
Size
1B

Context measured in tokens

Our take

Written Aug 2, 2026

Owsm CTC v3.2 ft 1B is a one-billion-parameter speech-to-text model from ESPnet that turns audio into written text. It is extremely fast and permissively licensed, though its accuracy drops sharply on difficult audio such as meetings and accented speech.

Who should pick it

Pick this for batch transcription of clean, prepared audio where speed matters most — it processes an hour of audio in about five seconds on benchmark hardware. Use it for projects that need attribution-only licensing, or for financial call transcription where it achieves low error rates. Skip it if you need hosted inference, are transcribing meetings or accented speech, or require verified non-English language support.

The case for it

  • Extremely fast: 692 times real time, processing an hour of audio in roughly five seconds on benchmark hardware.
  • Strong on clean, structured audio: about one word in forty-five wrong on read-aloud speech, and one in forty on financial calls.
  • Permissive Creative Commons Attribution 4.0 licence allows commercial use with attribution.

The case against it

  • Accuracy collapses in conversational and accented conditions: meetings are more than six times worse than read-aloud, and accented speech nearly three times worse than European-accented speech.
  • No hosted inference options available; you must self-host.
  • Unverified language coverage: all accuracy figures are English only, with no language list held.
00

How good is it?

TranscriptionTurning speech into text1.5 of 5Open ASR WER · 63rd of 74

Words it gets right

92.8%

Misses roughly one word in 14, averaged over nine English test sets.

How fast it listens

692×34th of 62

an hour of audio in 5 seconds, on the board's own hardware. Your machine will differ.

Where it struggles
Read aloudaudiobooks, clean recording2.2%
Podcasts and videoeveryday internet audio10.1%
Accented speechspeakers from many countries12.8%
Meetingsa room, several people, far microphone13.9%

Percentage of words wrong on each set, lower better. Bars are scaled to this model's own worst case, not to the board.

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.

Also scored, on boards we give no mark for
Financial calls 30th of 74Accented speech 56th of 74European-accented speech 60th of 74Recorded meetings 63rd of 74Harder read speech 63rd of 74Podcasts and video 65th of 74Clean read speech 67th of 74

These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done, which is why they get no rating.

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.
692.3independentsource ↗
7.3independentsource ↗
12.8independentsource ↗
2.5independentsource ↗
13.9independentsource ↗
10.1independentsource ↗
2.2independentsource ↗
4.8independentsource ↗
01

Can you run it yourself?

Fits in memory
weights load entirely on the card
Spills to system RAM
some weights offload; much slower
Too large
will not load even with offload
est
size is calculated; the verdict could change by 10%

Comfortable fit

A card many people ownFits in memory

GeForce RTX 4090 · 24 GB

Weights at Q4_K_M0.6 / 24 GBest
Spare memory20.9 GB spare
Usable context262Kwhat the spare memory holds; no published limit on record
Decode speed1330 tok/sest

Room to spare. 20.9 GB spare means a 10% error in the size would not change the answer.

One step upFits in memory

Apple M1 Pro (16-core GPU) · 32 GB

Weights at Q4_K_M0.6 / 32 GBest
Spare memory22.1 GB spare
Usable context262Kwhat the spare memory holds; no published limit on record
Decode speed231 tok/sest

Room to spare. 22.1 GB spare means a 10% error in the size would not change the answer.

On a MacFits in memory

Apple M2 (8-core GPU, 8GB unified) · 8 GB

Weights at Q4_K_M0.6 / 8 GBest
Spare memory4.1 GB spare
Usable context131Kwhat the spare memory holds; no published limit on record
Decode speed116 tok/sest

Room to spare. 4.1 GB spare means a 10% error in the size would not change the answer.

Your hardware
Checking your profile…

Memory use by level

Against a 24 GB card.

Q4_K_M
recommended
0.6 GBest
Fits in memory
Q5_K_M
0.7 GBest
Fits in memory
Q8_0
1.1 GBest
Fits in memory

All 3 sizes here are calculated, not measured. We hold no measured file for this model, so each size comes from the parameter count and every verdict above inherits that uncertainty.

Check against your own machine → · All 71 devices, with every size →

02

When we formed this view

Dates behind this page

Aug 2, 2026BenchmarkScored 12.8 on Accented speechleaderboard
Aug 2, 2026BenchmarkScored 2.5 on Financial callsleaderboard
Aug 2, 2026BenchmarkScored 13.9 on Recorded meetingsleaderboard
Aug 2, 2026BenchmarkScored 4.5 on European-accented speechleaderboard
Aug 2, 2026BenchmarkScored 10.1 on Podcasts and videoleaderboard
Aug 2, 2026BenchmarkScored 2.2 on Clean read speechleaderboard
Aug 2, 2026BenchmarkScored 4.8 on Harder read speechleaderboard
Aug 1, 2026ListedListed on LLMapfirst indexed by our pipeline
Aug 1, 2026BenchmarkScored 692.3 on Open ASR RTFxleaderboard
Aug 1, 2026BenchmarkScored 7.3 on Open ASR WERleaderboard

Prices last checked 6h ago

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.
03

Licence and identifiers

What the licence allowsCreative Commons Attribution 4.0, 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

Creative Commons Attribution 4.0

permissiveCommercial use allowed

Permissive content license: any use with attribution. Common for datasets and some model weights.

Identifiers

Modality record
audio->text
Catalogue slug
espnet-owsm-ctc-v3-2-ft-1b

Machine-readable model card (omc.json) →

Something wrong on this page? Tell us