How accurate is speech recognition in German?

German speech recognition is accurate with the right engine and noticeably weaker with Apple's built-in one. In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 4.84 percent ¹, Whisper large-v3-turbo had 5.08 percent ², and Apple SpeechTranscriber had 6.87 percent ³. That is the error count per 100 words on 100 German recordings from the FLEURS test set, with every engine running on the device.

Parakeet and Whisper are close, and both beat Apple's model

The two open engines are within a quarter of a point of each other, which on 100 recordings is too small a gap to call. Apple's model is clearly behind them. It makes up for it in speed: it ran about 72 times faster than real time ⁴, against about 20 times for Parakeet ⁵ and about 16 times for Whisper ⁶. For dictating a sentence at a time, all three are fast enough. See the speed dataset for every language.

Our measurement of all three engines in German

Lower is better in this table.

The transcripts behind each row are on the on-device benchmark page.

Published results put German between 3 and 5 percent for good models

On the same FLEURS German test, the Voxtral paper reports 3.38 percent ⁷ for Voxtral Small and 3.54 percent ⁸ for Voxtral Mini Transcribe. ElevenLabs Scribe scored 4.78 percent ⁹ and Whisper large-v3 scored 5.46 percent ¹⁰. NVIDIA reports 5.04 percent ¹¹ for Parakeet. Our on-device numbers match this spread.

The test set changes the result. On Multilingual LibriSpeech German, which is audiobook speech, ElevenLabs Scribe had 9.81 percent ¹², double its FLEURS figure.

Showing 8 of 31 rows. See every row in the dataset.

All published rows are on the published accuracy page.

German ranks in the middle of the leading languages

Taking the best engine in each language, German comes sixth lowest of the 25 languages in the benchmark. Italian (2.52 percent ¹⁹), Spanish (2.53 percent ²⁰) and English (3.92 percent ²¹) score lower, and so do Portuguese and Russian. German is well ahead of the Baltic and Balkan languages, where the best engines are several times worse.

One plausible reason German trails English is its long compound words: one misheard part makes the whole word wrong, and a different split counts as an error. The data does not test that explanation.

What to expect when you dictate in German

Choose Parakeet or Whisper if accuracy matters. Expect a handful of wrong words per 100, most often in compounds, names and loanwords, and capitalization of nouns is another place to watch. With Apple's built-in model expect clearly more corrections, though text appears faster. Mac dictation also depends on spoken punctuation: Apple's German help maps words such as "Komma" and "Punkt" to punctuation marks ²².

Read our guide to European languages in speech to text to pick an engine, and the languages overview for other languages.

How to choose an engine for German

If you dictate short messages, the speed of Apple's built-in model and its integration with the Mac make it the easy default, and you will correct more words. If you dictate reports or emails in German where every correction costs time, Parakeet or Whisper pay back their slower speed with fewer errors. Run a minute of your own typical speech through two engines before settling: your vocabulary, accent and microphone matter as much as the small gaps in the table above. The published accuracy page helps you judge how much a cloud service would add on top.

What these numbers do not show

The recordings are people reading prepared sentences in a quiet room, scored on one Mac. Dialects (Swiss and Austrian German), spontaneous speech, noise and technical vocabulary are not in the test. With 100 recordings per language, small gaps between engines are noise. For background on the measure, see the guide on how accurate speech recognition is. Compare with English and French.