TranscribePsy-250M-EU-sub-gguf

8-bit (Q8_0) GGUF build of TranscribePsy-250M-EU-sub for on-device speech recognition through the parakeet.cpp engine in our qvac-fabric-speech.cpp, in the patched release we publish on GitHub. The patches bound the memory of the mel front-end and add a Core ML decoder for the Apple Neural Engine; we are submitting them upstream as a pull request. The same engine runs the Core ML build.

This is the exact build used for every on-device measurement we publish. On six phones it averages a 0.052 real-time factor, 1,021 ms inference latency, a 252 ms cold start, and 534 MiB peak memory @ 10s, which makes it the fastest model on every device we tested.

See the base model card for accuracy, architecture, and training details.

Supported languages (8): Dutch (nl), English (en), French (fr), German (de), Italian (it), Polish (pl), Portuguese (pt), Spanish (es).

---

Which model should I use?

---

Files

---

Usage

# Build the engine (see BUILD.md in https://github.com/tether-ai-research/qvac-transcribePsy), then, from the repository root:
./build/parakeet --model transcribepsy-250m-eu-sub.gguf --wav audio.wav --threads 6

Audio must be 16 kHz mono WAV; the engine does not resample. ffmpeg -i input.m4a -ar 16000 -ac 1 audio.wav converts it.

Long recordings. Encoding a recording in one pass takes memory that grows markedly with its length. For recordings longer than a minute, use the command-line engine from our GitHub release: by default it encodes them in 60 s windows that overlap by 25 %, which kept peak memory below 2 GB on recordings of 10 to 30 minutes in our tests. Its README explains the settings, including a 30 s window for devices with less memory. In an app on a phone, set the window yourself, as shown in the README.

Download directly from the Hub:

hf download qvac/TranscribePsy-250M-EU-sub-gguf \
  transcribepsy-250m-eu-sub.gguf --local-dir .

---

On-device performance

Measured on six phones. The benchmark set is 160 utterances drawn at random from Multilingual LibriSpeech, 20 per language across eight languages, 40 minutes of audio in total. Every figure is the mean of the best three of five repeated runs, which keeps run-to-run thermal throttling out of the numbers.

On Apple silicon, the Core ML build runs on the Neural Engine and is faster still.

---

Limitations

• Eight languages only. Dutch, English, French, German, Italian, Polish, Portuguese, Spanish.
• Batch transcription, not streaming.
• Not intended for safety-critical or high-stakes automated decisions.

Citation

Model training and evaluation

@article{liu2026cafe,
  title={{CAFE}: Confining Feature Distillation to Language-Routed Adapters for Compact Multilingual Automatic Speech Recognition},
  author={Liu, Tianchi and Moratelli, Nicholas and Hashmi, Khurram Azeem and Zolfaghari, Mohammadreza and Jain, Rishabh and Lu, Louis and Park, Chang Dae and Nazir, Amril},
  journal={arXiv preprint arXiv:2610.19134},
  year={2026}
}

Technical report and on-device evaluation

Coming soon.

Copyright

We will take appropriate actions in response to notices of copyright infringement. If you believe your work has been used or copied in a manner that infringes upon your intellectual property rights, please email data-apps@tether.io identifying and describing both the copyrighted work and alleged infringing content.

Licensing

This model which was trained as described in the blog post (https://huggingface.co/blog/qvac/transcribepsy-250m)  is licensed by Tether Data, S.A. de C.V. under the Apache 2.0 license. As described in the blog post, this model is a version of the parakeet-tdt-0.6b-v3 (https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3) which is licensed under the CC-BY-4.0 license (Creative Commons Attribution 4.0 International).

As described in the blog post, the parakeet-tdt-0.6b-v3 model was used as a teacher model. 

The Yodas-granary dataset (https://huggingface.co/datasets/espnet/yodas-granary) is made available under the CC-BY-3.0 (Creative Commons Attribution 3.0 Unported). Yodas-granary is a subset of the Granary dataset (https://huggingface.co/datasets/nvidia/Granary) which is made available under the CC-BY-4.0 license (Creative Commons Attribution 4.0 International). The Voxpopuli dataset (https://github.com/facebookresearch/voxpopuli?tab=License-1-ov-file) is made available under the CC0-1.0 license (Creative Commons Zero v1.0 Universal). The Libri-light dataset (https://github.com/facebookresearch/libri-light) is made available under the MIT License (https://github.com/facebookresearch/libri-light?tab=MIT-1-ov-file). The Multilingual LibriSpeech (MLS) dataset (https://www.openslr.org/94/) is made available under the CC-BY-4.0 license (Creative Commons Attribution 4.0 International). The AMI Corpus dataset (https://groups.inf.ed.ac.uk/ami/corpus/) is made available under the CC-BY-4.0 license (Creative Commons Attribution 4.0 International). The LibriSpeech ASR Corpus (960 hours) dataset (https://www.openslr.org/12) is made available under the CC-BY-4.0 license (Creative Commons Attribution 4.0 International). The National Speech Corpus dataset (https://www.imda.gov.sg/how-we-can-help/national-speech-corpus) is made available under the Singapore Open Data Licence version 1.0 (https://data.gov.sg/open-data-licence). The VCTK Corpus (https://www.kaggle.com/datasets/pratt3000/vctk-corpus) is made available under the ODC-BY v1.0 (Open Data Commons Attribution License). The Europarl-ASR dataset (https://www.mllp.upv.es/git-pub/ggarces/Europarl-ASR) is made available under the CC-BY-4.0 license (Creative Commons Attribution 4.0 International) except that: (a) speech and text data from the European Parliament website (audio, official transcriptions and translations) incorporated into the Europarl-ASR dataset are the exclusive property of the European Union represented by the European Parliament and these data are reused here under the conditions stated European Parliament website's Legal notice (https://www.europarl.europa.eu/legal-notice/en/); (b) text data from the DCEP Digital Corpus of the European Parliament incorporated into the Europarl-ASR dataset are the exclusive property of the European Parliament and remain owned by the European Parliament and these data are reused here under the usage conditions of the DCEP Digital Corpus of the European Parliament (https://joint-research-centre.ec.europa.eu/language-technology-resources/dcep-digital-corpus-european-parliament_en#Usage%20Conditions); and (c) text data from the Europarl v10 corpus (https://www.statmt.org/europarl/) are reused here under the Europarl corpus terms of use. The Common Voice Scripted Speech 25.0 dataset (https://commonvoice.mozilla.org/en) is made available under the CC0-1.0 license (Creative Commons Zero v1.0 Universal). 

As described in the blog post, the parakeet-tdt-0.6b-v3 model was pruned and trained using the Yodas-granary, Voxpopuli, Libri-light, Multilingual LibriSpeech (MLS), National Speech Corpus, VCTK Corpus, AMI, LibriSpeech, and Europarl-ASR datasets and the following languages from the Common Voice Scripted Speech 25.0 dataset: bg (Bulgarian), cs (Czech), da (Danish), de (German), el (Greek), en (English), es (Spanish), et (Estonian), fi (Finnish), fr (French), hu (Hungarian), it (Italian), lt (Lithuanian), lv (Latvian), mt (Maltese), nl (Dutch), pl (Polish), pt (Portuguese), ro (Romanian), ru (Russian), sk (Slovak), sl (Slovenian), sv (Swedish), and uk (Ukrainian).