Hochdeutsch first, dialects with eyes open
Standard German (Hochdeutsch) — including regionally colored standard as spoken in Bavaria, northern Germany, or Vienna — is a top-tier language for the model. Compound words come out correctly assembled (Geschwindigkeitsbegrenzung, not three fragments), and noun capitalization follows German rules automatically.
The honest caveat: Swiss German (Schwiizerdütsch) is not Standard German. Alemannic dialect speech is effectively another language to the model — it will either normalize it roughly into High German or produce errors. Swiss speakers reading Standard German, or Swiss broadcast German, transcribe fine; a casual Zürich dinner conversation will not. The same applies, milder, to deep Bavarian and Austrian dialect — Austrian standard German is excellent, broad dialect is not.
German orthography, done properly
Output follows modern German spelling: capitalized nouns, umlauts (ä, ö, ü), ß where the reformed rules put it, and sentence punctuation suited to German's long clause structure. If a speaker uses English tech vocabulary mid-sentence — ubiquitous in German workplaces — those words are transcribed as English inline.
Frequently asked questions
Does Swiss German work?
Honestly: not well. Swiss German dialect (Schwiizerdütsch) differs from Standard German more than Dutch does, and the model treats it as noisy German — output tends to be a rough High-German normalization with real errors. Swiss speakers speaking Standard German transcribe excellently. For dialect recordings, budget editor time.
And Austrian German?
Austrian standard German — media, meetings, lectures — is top-tier, including Austrian vocabulary (Jänner, Erdapfel) written as spoken. Deep rural dialect gets rougher, the same pattern as Bavaria: the further from standard, the more the model normalizes or misses.
Are umlauts, ß and capitalized nouns handled?
Yes — the transcript is proper German orthography: ä/ö/ü, ß vs ss per the reformed rules, and nouns capitalized. You won't get "strasse" for "Straße" or lowercase nouns to fix afterwards.
How does it do with long compound words?
Well — compounds are where German speech models used to fall apart, and the large model handles them: Donaudampfschifffahrt-style coinages are assembled correctly when clearly spoken. Novel ad-hoc compounds in fast speech are the residual error source; they're quick edits.
Can I turn German audio into English text?
This page produces a German transcript — the accurate path. Export TXT and machine-translate it for an English version; two good steps beat one lossy one. In-app transcript translation is on the roadmap.
Denglisch — German with English mixed in?
Common in tech and business audio, and handled well: English terms ("das Meeting", "gebrieft", product names) are transcribed as English words inside the German sentence. Wholly English passages several sentences long transcribe better via the auto-detect uploader on the home page.