German audio and video to text

Hochdeutsch, Austrian, Swiss and regional speech

Standard German from Germany, Austria and German-speaking Switzerland is the target, and ordinary northern, southern and Austrian accents are generally manageable. Swiss German dialects (Schweizerdeutsch) are mostly spoken rather than written and differ from Hochdeutsch in vocabulary and grammar, so the model may return a loose Standard German rendering or simply get them wrong. Heavy Bavarian, Alemannic, Saxon or Plattdeutsch speech has the same problem, and interviews where people slide between dialect and Hochdeutsch come out unevenly.

Umlauts, ß, capital nouns and compounds

German text needs ä, ö, ü and ß, and every noun is capitalised, so glance at names and rare words when you proof-read. Long compounds such as Geschwindigkeitsbegrenzung are written as one word, which makes a wrongly split or joined compound a common edit. Swiss writing uses ss instead of ß, so do not treat that as an error in Swiss material. Formal Sie and informal du follow the speaker, and typographic German quotes („…“) are not guaranteed, so add them in your editor if your style needs them.

What speakers of German use it for

Students record Vorlesungen and Seminare to turn them into study notes. Journalists and researchers transcribe interviews for qualitative work. Teams in German-speaking offices record Besprechungen and video calls to write minutes. Creators add German captions to YouTube videos and short clips. Organisations with data protection concerns (the DSGVO question comes up often) can read about the self-hosted option, where audio stays on their own server.

Which model works best for German

German is one of the languages Whisper-family models have seen the most speech in, so the small model can give a reasonable first draft from clear audio. That is not a promise of accuracy: it still slips on names, numbers, jargon and overlapping or noisy speech, so proof-read anything you will publish or rely on.

On tryscriptorium.com the no-account preview and regular accounts run the small model (the larger medium and large-v3 models run when you self-host Scriptorium).

German has plenty of training data, so the small model is a reasonable starting point for clear, single-speaker recordings. Expect more mistakes with Swiss German, thick dialect, overlapping speakers and specialist vocabulary. Proof-read names, numbers and compound words.

Subtitles and captions in German

German puts the verb late and compounds run long, so lines can get wide. Aim for two lines of roughly 40 to 42 characters and split at clause boundaries rather than mid-phrase. Exports are UTF-8, so umlauts and ß display correctly in VLC, YouTube and most editors. Subtitles stay German, the language shown in the transcript.

The uploader is set to German

The upload box on this page transcribes as German, so the recogniser does not have to guess the language from the first seconds. If the recording is in another language or switches languages, use Audio to text, which detects the language automatically. Automatic transcription is a draft and no accuracy figure is promised: check names, numbers and anything you will rely on.