How accurate is speech recognition in Spanish?

Spanish is one of the easiest languages for speech recognition: the best on-device engines make fewer than 3 wrong words per 100 on read speech. In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 2.53 percent ¹, Parakeet TDT 0.6B v3 had 2.79 percent ², and Apple SpeechTranscriber had 4.30 percent ³. The test was 100 Spanish recordings from the FLEURS set, all run on the device.

Whisper and Parakeet are nearly tied, and Apple's model trails on accuracy

The two open engines differ by a quarter of a point, which on 100 recordings is within noise. Apple's model has a clearly higher error rate but is the fastest: about 57 times faster than real time ⁴, against about 22 times for Parakeet ⁵ and about 15 times for Whisper ⁶. The speed dataset lists every language.

Our measurement of all three engines in Spanish

Lower is better in this table.

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

Published results put Spanish at 2.5 to 3.5 percent on FLEURS

On FLEURS Spanish, the Voxtral paper reports 2.58 percent ⁷ for GPT-4o mini Transcribe, 2.72 percent ⁸ for Voxtral Small and 2.81 percent ⁹ for Whisper large-v3. NVIDIA reports 3.45 percent ¹⁰ for Parakeet. Our on-device results fall in the same band.

Harder audio raises the figure. On Multilingual LibriSpeech Spanish, ElevenLabs Scribe had 8.97 percent ¹¹, nearly three times its FLEURS result of 3.13 percent ¹².

Showing 8 of 32 rows. See every row in the dataset.

All published rows are on the published accuracy page.

Spanish ranks second of 25 languages by its best engine

Taking the best engine in each language, Spanish has the second lowest error rate of the 25 languages in the benchmark, a hair behind Italian (2.52 percent with Parakeet ¹⁷). English (3.92 percent ¹⁸) and French (5.29 percent ¹⁹) are less accurate. Spanish spelling follows pronunciation closely, which plausibly helps, though the benchmark does not test that.

What to expect when you dictate in Spanish

Expect around 3 wrong words per 100 with Whisper or Parakeet in a quiet room, and a few more with Apple's built-in model. Errors cluster in proper names, foreign terms and numbers. Accents and regional vocabulary matter: Apple lists five Spanish variants for on-device dictation on the Mac ²⁰, so pick the one that matches how you speak. Apple's Spanish command table has no spoken command for the opening question mark "¿" ²¹, so plan to type it or fix it afterward.

Our guide to European languages in speech to text helps you choose an engine, and the languages overview covers other languages.

How to choose an engine for Spanish

Because Spanish error rates are low for the open engines, the choice comes down to what you value. Whisper and Parakeet give the fewest corrections. Apple's model is the quickest and already built into the Mac, which suits short messages. If your Spanish includes a lot of English product names, test with your own sentences before you decide, since mixed-language speech is not in this benchmark. Speak a minute of your usual text into two engines and count the corrections: your accent, vocabulary and microphone will matter more than the small gaps in the table above. The published accuracy page shows how cloud services compare.

What these numbers do not show

The recordings are people reading prepared sentences in a quiet room, scored on one Mac. The test does not separate Spain from Latin America or Caribbean speech, and it has no spontaneous speech, noise or code-switching with English. With 100 recordings, small gaps are noise. For the measure itself, see the guide on how accurate speech recognition is. Compare with English and Italian.