How accurate is speech recognition in Maltese?

Maltese is the hardest language in our benchmark, and only one engine can be used for it. In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 22.02 percent ¹ on 100 Maltese recordings from the FLEURS test set, while Whisper large-v3-turbo had 72.35 percent ². We have no Apple SpeechTranscriber result for Maltese.

Whisper gets about three words in four wrong in Maltese

A word error rate above 70 percent means the transcript is closer to a rough guess than to a draft. Parakeet makes about a third of Whisper's errors. It is also the faster engine, at about 21 times faster than real time ³, against about 13 times for Whisper ⁴. The speed dataset covers every language.

Our measurement of the engines in Maltese

Lower is better in this table.

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

Published results confirm the gap

NVIDIA reports 20.46 percent for Parakeet on FLEURS Maltese ⁵, close to our 22.02 percent. OpenAI's paper gives 76.6 percent for the older Whisper large-v2 ⁶, close to our turbo result. Two independent sets of measurements agree: Whisper has not learned Maltese well, and Parakeet has.

All published rows are on the published accuracy page.

Maltese ranks last of 25 languages, even with Parakeet

Taking the best engine in each language, Maltese is twenty-fifth of the 25 in the benchmark. Its best result is still worse than Slovenian (19.83 percent with Parakeet ⁷) and Lithuanian (17.75 percent ⁸), and many times worse than Italian (2.52 percent ⁹). Maltese is spoken by a small population, so models have less training material, and it mixes Semitic roots with Romance and English loanwords.

Formatting costs Parakeet four points in Maltese

Counting punctuation and capital letters as a reader sees them, Parakeet's text differed from the written sentences by 25.96 errors per 100 tokens ¹⁰, and Whisper's by 71.96 ¹¹. For Whisper the two figures are nearly the same, because so many words are wrong that formatting hardly matters.

What to expect when you dictate in Maltese

With Parakeet in a quiet room, expect roughly 1 wrong word in 5, so dictation works as a first draft that you must edit closely. It can still be faster than typing a long text if you correct as you go, but short messages are likely quicker by keyboard. Errors are most likely on names, English words mixed into Maltese and the special letters ċ, ġ, ħ and ż. Do not use Whisper for Maltese. If your dictation app does not offer Parakeet, it is not a good fit for Maltese. We found no published fact that confirms Maltese as a built-in Mac or Windows dictation language, so check your system's language list.

An app must use Parakeet to handle Maltese at all

Most languages in the benchmark have two engines of similar quality, so the choice hardly matters. Maltese does not. An app that uses Whisper for every language will serve Maltese badly, and one that uses Parakeet will serve it acceptably. Check which engine your dictation app uses before you rely on it, and test it with a page of your usual text. The published accuracy page shows how other models compare on similar tests.

What these numbers do not show

The recordings are people reading prepared sentences in a quiet room, scored on one Mac with 100 recordings. Spontaneous speech, code-switching with English, noise and domain vocabulary are not in the test, and we did not measure Apple's model. For the measure itself, see the guide on how accurate speech recognition is.

Our guide to European languages in speech to text helps you choose an engine, and the languages overview covers other languages. For comparison, read the same question for Greek, Slovenian and Italian.