Models / Sounds Good AI/ Zipformer cr CTC Transducer XL 290M

Zipformer cr CTC Transducer XL 290M

Sounds Good AI · released Jul 9, 2026 · soundsgoodai/Zipformer-cr-ctc-transducer-XL-290M

Speech to textTranscribes a recording into words

Input: audio. Output: text.InputOutput
Type
Open weightsCreative Commons Attribution-NonCommercial 4.0
Languages
1
Size
0.3B

Context measured in tokens

Our take

Written Aug 2, 2026

Zipformer cr CTC Transducer XL is a tiny downloadable speech-to-text model that turns English audio into written words at roughly 160 times real time. It is nearly flawless on clean read-aloud recordings but its accuracy collapses in meetings, podcasts and accented speech.

Who should pick it

Pick this for batch transcription of clean, read-aloud English audio where speed matters — it processes an hour of audio in about twenty-three seconds on benchmark hardware. Use it for local deployment on very constrained devices, or non-commercial projects where hardware cost must be negligible. Skip it if you need commercial licensing, any language other than English, or reliable accuracy on unscripted or accented speech.

The case for it

  • Extremely fast: roughly 160 times real time on benchmark hardware.
  • Nearly flawless on clean read-aloud audio, with about one word in seventy-six wrong.
  • Tiny enough to run almost anywhere at 0.29B parameters, with no token-billing constraints.
  • Permissive for non-commercial use under a Creative Commons licence.

The case against it

  • Accuracy degrades sharply in natural settings: more than seven times the error rate in recorded meetings compared with clean speech, and nearly six times on podcasts and video.
  • No commercial use permitted, and no hosted options available in our catalogue.
  • English only; no other language is supported.
00

How good is it?

TranscriptionTurning speech into text4 of 5Open ASR WER · 23rd of 74

Words it gets right

94.8%

Misses roughly one word in 19, averaged over nine English test sets.

How fast it listens

160×51st of 62

an hour of audio in 23 seconds, on the board's own hardware. Your machine will differ.

Languages

1

Listed on the model card. The accuracy above is English only.

Where it struggles
Read aloudaudiobooks, clean recording1.3%
Podcasts and videoeveryday internet audio8.3%
Accented speechspeakers from many countries7.6%
Meetingsa room, several people, far microphone10.3%

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
Accented speech 2nd of 74Financial calls 4th of 74Clean read speech 32nd of 74Harder read speech 32nd of 74Recorded meetings 35th of 74Podcasts and video 39th of 74European-accented speech 55th 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.
159.8independentsource ↗
5.2independentsource ↗
7.6independentsource ↗
1.6independentsource ↗
10.3independentsource ↗
8.3independentsource ↗
1.3independentsource ↗
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_M0.2 / 24 GBest
Spare memory21.4 GB spare
Usable context262Kwhat the spare memory holds; no published limit on record
Decode speed4587 tok/sest

Room to spare. 21.4 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_M0.2 / 32 GBest
Spare memory22.6 GB spare
Usable context262Kwhat the spare memory holds; no published limit on record
Decode speed796 tok/sest

Room to spare. 22.6 GB spare means a 10% error in the size would not change the answer.

On a MacFits in memory

Apple M1 (8-core GPU, 8GB unified) · 8 GB

Weights at Q4_K_M0.2 / 8 GBest
Spare memory4.6 GB spare
Usable context262Kwhat the spare memory holds; no published limit on record
Decode speed271 tok/sest

Room to spare. 4.6 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
0.2 GBest
Fits in memory
Q5_K_M
0.2 GBest
Fits in memory
Q8_0
0.3 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 7.6 on Accented speechleaderboard
Aug 2, 2026BenchmarkScored 1.6 on Financial callsleaderboard
Aug 2, 2026BenchmarkScored 10.3 on Recorded meetingsleaderboard
Aug 2, 2026BenchmarkScored 4.3 on European-accented speechleaderboard
Aug 2, 2026BenchmarkScored 8.3 on Podcasts and videoleaderboard
Aug 2, 2026BenchmarkScored 1.3 on Clean read speechleaderboard
Aug 2, 2026BenchmarkScored 3 on Harder read speechleaderboard
Aug 1, 2026ListedListed on LLMapfirst indexed by our pipeline
Aug 1, 2026BenchmarkScored 159.8 on Open ASR RTFxleaderboard
Aug 1, 2026BenchmarkScored 5.2 on Open ASR WERleaderboard

Prices last checked 7h 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-NonCommercial 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-NonCommercial 4.0

restricted_openNon-commercial

Weights are downloadable but commercial use is prohibited. Research and personal use only.

Identifiers

Modality record
audio->text
Catalogue slug
soundsgoodai-zipformer-cr-ctc-transducer-xl-290m

Machine-readable model card (omc.json) →

Something wrong on this page? Tell us