Is speech to text as accurate in Spanish as in English?

In the one published benchmark we found that tests both languages, Spanish speech recognition scored better than English, not worse. That benchmark uses read speech and a single research model, so it does not measure the dictation app on your phone or computer. Accuracy also varies by Spanish dialect, so the right answer for you depends on the variant you speak and how you speak it.

On a read-speech benchmark, Spanish scored better than English

The Fleurs benchmark measures speech recognition on read sentences. In it, Whisper large-v2 had a word error rate of 3.0 percent for Spanish ¹, and 4.2 percent for English ². That makes the English error rate about 1.4 times the Spanish rate ³.

The word error rate counts the words a recognizer gets wrong, as a share of the words spoken. A lower figure is better. The benchmark's conditions are narrow, and the table shows where everyday Spanish dictation differs.

The paper tests one model, not the built-in dictation on any particular device. Apple and Microsoft do not publish a Spanish error rate for their dictation in the sources we checked, so we cannot say how they compare with Whisper. The Fleurs result tells you that Spanish is a language these systems handle well. It does not tell you what you will get on a noisy train.

Accuracy changes by dialect, not just by language

Language is not the only factor. A 2026 preprint tested YouTube's automatic Spanish captions on recordings from seven Latin American dialects, and it found systematic differences that it attributes to specific dialects ⁴. Mexican Spanish had an overall word error rate of 18 percent ⁵, and Argentinian Spanish had the highest rate in that test, at 24 percent ⁶.

This study is weaker evidence for dictation than the Fleurs benchmark. It tests a caption system, not a dictation app. Each dialect came from a different set of recordings, so a gap between two dialects may partly reflect the recordings. It also covers Latin American Spanish only. Still, it shows why a single "Spanish accuracy" figure can hide large differences. Our guide to Spanish variants explains how to choose a variant for your speech.

Missing accents count as errors, and they add up

Spanish accents are part of the spelling. A word written without its tilde counts as a wrong word in a word error rate, even when the recognizer heard the sound correctly. A 2016 study from Costa Rica describes missing tildes as one of the most common errors of Spanish speech recognition ⁷, so an accurate-sounding draft can still have many errors. The accent mark question covers the checks to make.

Mixed speech also raises the error rate. If you move between Spanish and English within one sentence, the recognizer may write the wrong language for some words. The code-switching entry describes the problem, and the multilingual dictation guide explains the settings.

It depends on your speech, your variant and your environment

• Read speech is easier than spontaneous speech. Your dictation is probably less tidy than a read passage, so expect more errors than the benchmark shows.
• Your variant matters. A variant that does not match your speech can add errors, so test the one you pick.
• Noise matters. A quiet room is the closest to the benchmark, and a noisy room is further from it.
• The app matters. Each product uses its own recognizer, and the Fleurs figures do not transfer to them.

How to measure the accuracy of your own dictation

The test below takes about ten minutes, and it gives you a figure for your voice, in your room, with your app.

1. Read aloud a 100-word passage you wrote, in Spanish, in the setting where you normally dictate.
2. Count the words the recognizer got wrong. Count a word with a missing accent as wrong, and count each missing ¿ or ¡ as one error.
3. Divide the wrong words by 100 to get your error rate for that passage.
4. Repeat the test in English with a passage of the same length, so you can compare the two.
5. Run the test again in a noisier room, or with a different variant, to see what changes.

Compare your numbers with each other, not with the benchmark. A result that is much worse in Spanish than English points to the variant, the vocabulary or your accent marks, and you can fix each one.

For the full guide, see the Spanish speech to text guide. The question on whether speech recognition is worse in other languages covers the same question across languages, and the talk to text question for English covers translation.