What a computer can do with Swiss German
And what it cannot. This page collects what has actually been measured in the field — with the numbers and the sources, so you can check our claims against them.
Why it is hard
There is no official spelling. There are recommendations from 1938 that dialectology uses — but even trained transcribers apply them differently, and almost nobody writes that way to a friend.
Dialect is spoken; the standard is written. So writing down what was said is not transcription here, it is translation — and that is how nearly every system that exists is built.
And it is a small language in the sense that matters for data: the largest public collections are a few hundred hours, and almost all of them are licensed for research only.
Where the data comes from
The public collections this field rests on. The direction column is the one to read: it shows that almost everything hears dialect and writes the standard.
ArchiMob 2016
dialect, written
- licence
- CC BY-NC-SA 4.0 · text; audio on request
Swiss Parliaments Corpus 2021
dialect heard → standard written
- hours
- 293
- speakers
- 198
- licence
- MIT
SwissDial 2021
text → dialect spoken
- regions
- 8
- licence
- no licence published
SDS-200 2022
dialect heard → standard written
- hours
- 200
- speakers
- 3,816
- licence
- research only
STT4SG-350 2023
dialect heard → standard written
- hours
- 343
- speakers
- 316
- regions
- 7
- licence
- research only
Understanding
Word error rate on the same test set, so the numbers can be compared. All of these produce Standard German — the figure says how well it translated, not how well it wrote dialect.
23%
word error rate
14%
word error rate
XLS-R 1B 2023
fine-tunedweights published
12.1%
word error rate
Whisper large-v2 2025
fine-tunedweights not published
Speaking
Here the marketplace misleads. What is sold as a Swiss German voice is usually Swiss Standard German — the written language, read aloud. Real dialect synthesis exists almost only in research.
Commercial de-CH voices
Swiss Standard Germanservice
ETH Zurich, Swiss Voice
dialectresearch
T5 and VITS research pipeline 2023
dialectresearch
Voice cloning from podcasts 2025
dialectresearch
Language models
Whether a model really handles dialect, or whether that is only in the press release. Evaluated means somebody measured it and published the result.
SwissBERT 2023
dialect not evaluatedCC BY-NC 4.0
SwissBERT + gsw 2024
dialect evaluatedCC BY-NC 4.0
Apertus 2025
dialect not evaluatedApache 2.0
What this means for Heidi
Dictation does not write dialect down. It writes what you want to say, in the language you already have — which is exactly what the research can do.
Heidi does not speak. A voice that pronounced Zurich German wrongly is something you could not check, and that is the one mistake this product must not make.
The dialect check runs without a model. It is a fixed list of rules, not a language model, which is why it cannot start inventing things.
Sources
- Samardžić, Scherrer & Glaser 2016. ArchiMob — A Corpus of Spoken Swiss German. Proceedings of LREC 2016, 4061–4066, Portorož.
- Plüss, Neukom, Scheller & Vogel 2021. Swiss Parliaments Corpus, an Automatically Aligned Swiss German Speech to Standard German Text Corpus. Proceedings of the Swiss Text Analytics Conference 2021, CEUR-WS Vol-2957.
- Dogan-Schönberger, Mäder & Hofmann 2021. SwissDial: Parallel Multidialectal Corpus of Spoken Swiss German. arXiv preprint 2103.11401 — not peer-reviewed.
- Plüss, Hürlimann, Cuny, Stöckli, Kapotis, Hartmann, Ulasik, Scheller, Schraner, Jain, Deriu, Cieliebak & Vogel 2022. SDS-200: A Swiss German Speech to Standard German Text Corpus. Proceedings of LREC 2022, 3250–3256, Marseille.
- Plüss, Deriu, Schraner, Paonessa, Hartmann, Schmidt, Scheller, Hürlimann, Samardžić, Vogel & Cieliebak 2023. STT4SG-350: A Speech Corpus for All Swiss German Dialect Regions. Proceedings of ACL 2023 (Short Papers), 1763–1772, Toronto.
- Dolev, Lutz & Aepli 2024. Does Whisper Understand Swiss German? An Automatic, Qualitative and Human Evaluation. Proceedings of VarDial 2024, 28–40, Mexico City.
- Timmel, Paonessa, Vogel, Perruchoud & Kakooee 2025. Fine-tuning Whisper on Low-Resource Languages for Real-World Applications. Proceedings of the Swiss Text Analytics Conference 2025.
- Bollinger, Deriu & Vogel 2023. Text-to-Speech Pipeline for Swiss German — A Comparison. arXiv preprint 2305.19750 — not peer-reviewed.
- Stucki, Deriu & Cieliebak 2025. Voice Adaptation for Swiss German. arXiv preprint 2505.22054 — submitted, not yet accepted.
- Vamvas, Graën & Sennrich 2023. SwissBERT: The Multilingual Language Model for Switzerland. Proceedings of the Swiss Text Analytics Conference 2023, 54–69.
- Vamvas, Aepli & Sennrich 2024. Modular Adaptation of Multilingual Encoders to Written Swiss German Dialect. Proceedings of the 1st Workshop on Modular and Open Multilingual NLP, 16–23, St Julians.
- Project Apertus (EPFL, ETH Zurich & CSCS) 2025. Apertus: Democratizing Open and Compliant LLMs for Global Language Environments. arXiv preprint 2509.14233. Swiss German appears as 6,000 post-training instruction examples; no dialect evaluation is reported.
- Dieth 1986. Schwyzertütschi Dialäktschrift. 2nd edition, Sauerländer, Aarau; first published 1938 as recommendations of a commission of the Neue Helvetische Gesellschaft. Widely used in dialectology, applied inconsistently even within one corpus, and unknown to most writers — a proposal, not an orthographic standard.
- Zampieri, Malmasi, Scherrer, Samardžić, Tyers, Silfverberg, Klyueva, Pan, Huang, Ionescu, Butnaru & Jauhiainen 2019. A Report on the Third VarDial Evaluation Campaign. Proceedings of VarDial 2019, 1–16, Ann Arbor. Best macro F1 on four Swiss German dialect areas: 0.76.