How accurate is speech recognition in Danish?

Danish is one of the harder European languages for on-device speech recognition. On 100 read FLEURS sentences, Whisper large-v3-turbo had a word error rate of 14.57 percent ¹ and Parakeet TDT v3 had 15.78 percent ². Roughly one word in seven comes out wrong, even in a quiet room and with clean read speech.

Whisper is the more accurate of the two engines on Danish, by about one point. That is the opposite of the picture in most other languages we tested, where Parakeet is equal or better.

Whisper edges Parakeet on Danish, but both miss about one word in seven

Both engines are close. Parakeet is the faster one: it transcribed the Danish recordings 21 times faster than real time ³, while Whisper managed 13 times faster than real time ⁴. Apple SpeechTranscriber has no row here: in our benchmark it ran only on English, German, French, Spanish, Italian and Portuguese, so for Danish the choice is between two engines.

If you also count punctuation and capital letters as a reader sees them, the gap to the written sentences grows. Whisper's formatted error rate was 16.90 percent ⁵ and Parakeet's was 18.54 percent ⁶. Dictated Danish needs more cleanup than the word figure alone suggests.

Published results rank Danish low too

Published word error rates tell the same story. On the FLEURS Danish test, NVIDIA's own figure for Parakeet TDT 0.6B v3 is 18.41 percent ⁷, close to our 15.78 percent ². For an older and larger model, the Whisper paper reports 13.8 percent ⁸ for Whisper large-v2 on Danish. These are the same test set but not the same recordings or engine versions, so compare the pattern, not the decimals.

Danish sits near the bottom of the 25 languages

Compared with the other languages in the benchmark, Danish is in the weaker third. Italian reached 2.52 percent ⁹ with Parakeet and Dutch 6.32 percent ¹⁰. Estonian (14.17 percent ¹¹) and Hungarian (14.09 percent ¹²) are close to Danish with Parakeet, and Latvian (17.27 percent ¹³) is further behind. Danish is not a hopeless case, but it is clearly harder than its neighbors Dutch and Swedish (9.95 percent ¹⁴ with Whisper).

What to expect when you dictate in Danish

For a short message, expect to fix a word or two per sentence of about fifteen words. Names, compound words and numbers are where errors cluster in most languages, so read your text back before sending anything important.

Choose Whisper large-v3-turbo if accuracy on Danish is all that matters and a slower engine is fine. Choose Parakeet if you value speed: its speed advantage is large and the accuracy cost on Danish is small. Yap, which we make, uses Parakeet. In both cases the text stays on your Mac, because both engines ran with nothing sent over the network.

How to read these numbers for Danish

A word error rate counts every substituted, deleted or inserted word against the written sentence, so a single wrong ending on a long word costs a full word. This benchmark does not tell us why Danish is harder, only that it is. If a text matters, dictate in short phrases, keep a steady pace, and review names and numbers by eye. The gap between the two engines is small enough that your voice, microphone and subject matter will matter more than the engine.

What the data does not show

The recordings are read sentences from Wikipedia, spoken in a quiet room, and every engine heard the same 100 of them on one Mac in one run. Real dictation has hesitations, names, numbers and background noise, and none of that is in this benchmark. Numbers and spelling variants are not normalized, so "25" against "twenty-five" counts as an error for every engine alike. Treat the gaps of a point or two between engines as small, and the gaps of five points or more as real.

Where to read more

• All languages in one place: dictation languages, accents and translation and the guide to speech to text in European languages.
• The full rows and every transcript: the on-device ASR benchmark, the published word error rates and the speed data.
• Sibling pages: Dutch, Polish, Czech.