How do you transcribe Spanish, German or Dutch audio to text?

Set the spoken language to Spanish, German or Dutch, transcribe the recording with any tool that supports it, and read the result against the audio. In a published test on read sentences, OpenAI's Whisper large-v2 had a word error rate of 3.0 percent in Spanish ¹, 4.5 percent in German ² and 6.7 percent in Dutch ³. Those are best-case figures. Accent, region and the sound quality of the recording still change the result, so the check after transcription matters as much as the choice of tool.

This page covers the steps, the cases where the answer changes, and the checks to run. For the wider method, start with the guide to transcribing audio in other languages.

The three figures come from read speech, which is a best case

The three figures above are results for read sentences in a benchmark, not for dictation at your desk. Real dictation, accents and background noise can raise error rates, so a result for Spanish, German or Dutch in one test does not predict your own recording. The guide to transcribing audio in other languages covers the wider picture across languages.

Read-speech numbers are a best case, and conversation is usually harder. The better guide to a real recording is a short test on that recording, described below.

Spanish accuracy depends on the dialect

Spanish is spoken in many countries with different accents and vocabulary, and the results vary with them. In a 2026 test of YouTube's automatic Spanish captions, which is not the same system as Whisper, Argentinian Spanish had the highest overall word error rate of the dialects tested, at 24 percent ⁴. That gap is a reminder that a result for "Spanish" is really a result for the varieties that were tested.

The practical lesson is to match the regional setting to the speaker when the tool offers one. Choose the variant that fits the speaker's country, and then test a recording from the same speaker before you rely on it. For how Spanish dictation handles accent marks and punctuation, see the question on which Spanish to use for dictation.

German has a dialect problem and a word problem

Many tools treat German as one language, but many German speakers talk in dialect. Swiss German is the clearest case. Dolev and colleagues tested Whisper large-v3 on Swiss German with no training on Swiss German. Scored against Standard German reference text, it reached a word error rate of 23.0 percent on their test set ⁵. If you need the dialect as it is spoken, plan for an error rate at that level or higher, and a human check.

German also builds long words from parts. In a 2018 study, Milde and Köhn say that German is a very productive language, so compounds unknown to the language model are quite frequent ⁶. A compound the model has never seen is therefore a likely source of error. Read compound nouns against the audio, especially technical and legal terms.

Dutch can differ by region, and the tool may list only one version

Dutch is spoken in both the Netherlands and Belgium, but the tools do not always list both. Windows 11 voice typing lists Dutch (Netherlands) as its only Dutch entry, with no Belgian Dutch entry ⁷. If your speaker is Belgian and the tool offers only the Netherlands entry, test the output before you use it. Regional words and place names are the places to check first.

Set the language yourself and check the start of the transcript

Set the language before you transcribe whenever the tool lets you. Automatic detection is a guess, and a wrong guess can affect the whole transcript. Transcribe the first minute or two and read it. If the words look like another language, change the setting and run it again. Then check the rest of the transcript at several points, not only at the start, because errors often appear later when the speaker changes topic or speed.

If a recording includes English words, names or brand names, check that they came out spelled as they were said, and not respelled in the language's form.

Recording a conversation in Germany has legal limits

Before you record other people, check the rules where you are. Germany's Criminal Code, Section 201, makes it an offence, punishable by up to three years in prison or a fine, to record the non-public spoken words of another person without authorization ⁸. This is a general description, not legal advice. In practice, ask the speakers' permission before you record a private conversation, and then transcribe it.

What to do, step by step

1. Set the language and the regional variant. Choose Spanish, German or Dutch, and pick the entry closest to the speaker's country or region.
2. Transcribe the first minute or two and read it against the audio.
3. Test on your own audio. Choose three to five recordings, including the hardest ones: fast speech, background noise and any dialect. Correct one by hand, and count the errors to get your own word error rate.
4. Proofread the whole transcript at several points, paying most attention to names, numbers, dates and compound nouns.
5. Decide what an error would cost. For a legal, medical or published transcript, plan a human review.

If you want to dictate your own text rather than transcribe a recording, Yap, which we make, is an on-device dictation app for Mac and Windows, and the download page is the place to check what it supports. For a file you upload in a browser, our free audio to text tool transcribes it in the browser. Check its output with the steps above.

Other questions in this guide: Arabic or Hebrew audio, Tamil, Urdu or Bengali audio, Hindi audio, and translating a voice recording to English. For the wider method, see how to transcribe audio to text and the guide to choosing transcription software.