Dictation Languages: What Each App and Device Supports

Major dictation tools now handle many languages, but their lists do not match. Microsoft's voice typing in Windows 11 lists 46 languages and locales ¹. The main list for Dictate in Microsoft Word has 15 entries ². Google's cloud speech service says it supports over 125 languages ³. These figures count different things, so compare them with care. Before you pick a tool, check your exact language and region on that vendor's own list, whether it works without a connection, and whether punctuation and commands work in your language.

This guide is part of our overview of dictation languages. It puts the lists side by side, marks which tools need a connection, and explains why a supported language is not always an accurate one. It links to the question pages for the common searches, and to the guide on changing your dictation language for the setup steps.

Major tools cover many languages, but no two vendors count them the same way

The counts measure different things. Apple and Google publish lists that depend on the device and the software version. Microsoft publishes a list for each product. Google's cloud service is an API that developers call, not a dictation app. Whisper is an open speech model, and its count is the number of languages in its training data ⁴, not a list of languages a product offers. The chart below puts the stated numbers next to each other, and the table after it explains what each one covers.

Table 1. What each tool states, and what the number counts

Read each row as a starting point. A large number tells you the tool has been trained or tested on many languages. It does not tell you that every one of them works well on your computer, in your app, or in your accent. The rest of this guide covers those gaps.

Apple, Microsoft and Google all cover the big languages and differ on regional variants

All of the major tools cover the widely spoken languages, so most readers who dictate in English, Spanish, French, German, Chinese, Japanese or Hindi will find a match. The differences appear at the level of region. Apple's iPhone list is organized by language and region, so English appears several times for different countries ⁷. On a Mac, Apple's list includes both Mandarin Chinese (China mainland) and Mandarin (Taiwan) ¹¹, which are separate choices for speakers of different varieties.

Microsoft's Word list follows the same pattern. Its main list has five English entries, for Australia, Canada, India, the United Kingdom and the United States, and two Spanish entries, for Spain and Mexico ². A language name on a list can therefore hide several choices, and the right choice depends on where you live.

Regional gaps can also appear between Microsoft's own products. Microsoft's Azure Speech table includes Bengali (India) and Urdu (India), but Windows voice typing does not offer them ¹². Someone who dictates in Bengali on Windows will need another tool, even though the same company supports those languages in its speech service. The Windows guide to dictation languages lists the Windows locales in more detail, and the Mac guide to dictation languages covers Apple's Mac list.

Offline dictation is set per language, so a language can work online and fail offline

Two of the tools in Table 1 need a connection for every language. Windows voice typing needs an internet connection, a working microphone and the cursor in a text box before it will type what you say ⁵. Microsoft says the feature uses online speech recognition powered by Azure Speech services ¹³. Dictate in Word also needs a microphone and a reliable internet connection ⁶. For these two tools, a supported language on the list does nothing for you when you are offline.

Apple takes a different approach. On iPad, dictation is processed on the device in many languages and needs no internet connection ⁸. The phrase "many languages" matters: Apple does not say which languages work without a connection, so check your own language on your own device.

Google's history shows how offline support arrives one language at a time. In 2019 Google moved Gboard voice typing onto the phone with an 80 MB speech recognizer that works offline ¹⁰. The first on-device recognizer launched on Pixel phones in American English only ¹⁴. Chromium's source for ChromeOS dictation tracks, for each locale, whether offline recognition is supported and whether the matching language pack is installed ¹⁵.

For practical purposes, check two things for your language before you rely on a tool: whether it has an offline mode for that language, and whether the language pack is installed. If you dictate on a train, on a plane or in a place with weak signal, test the offline mode before you need it. The glossary entry on on-device speech recognition explains what runs locally and what does not.

Dragon's list depends on the edition, while Whisper's training set covers 99 languages

We could not confirm a full, current language list for Dragon NaturallySpeaking. Nuance's Dragon Professional page describes the product's accuracy and features but does not publish a language list at dragon.nuance.com/dragon-professional. Check the edition you plan to buy before you rely on a language, and ask the vendor to confirm it in writing if you need it for work. Our overview of Dragon NaturallySpeaking covers what the product does, and the question on Dragon's languages covers the search answer.

Whisper is different because its design is public. The OpenAI model predicts the spoken language with one token for each language in its training set, and the paper counts 99 tokens in total ⁴. Whisper can transcribe and translate, and you can run it on your own computer. OpenAI's README says that Whisper's performance "varies widely depending on the language" ¹⁶, and it publishes a per-language chart of error rates in the model's documentation.

A language token means the model saw training data in that language. It does not mean the model is accurate in that language. Some languages are missing from the training data altogether. The authors of one 2025 study say Whisper's training data includes no isiXhosa and no closely related languages, so they used Swahili as the language setting when they fine-tuned on isiXhosa ¹⁷. For more on the model itself, see the guide to Whisper's supported languages and the overview of Whisper.

A supported language is not always an accurate one: low-resource languages lag far behind

A supported language is one the tool can recognize at all. It is not a promise of the same accuracy as English. The clearest number comes from FLEURS, a public speech benchmark that covers 102 languages ¹⁸. On that benchmark, Whisper large-v2's word error rate in Yoruba was about 31.6 times its rate in Spanish ¹⁹. The same model is usable in Spanish and far weaker in Yoruba, and the list alone would not show you that.

Vendors tend to report results for their largest languages. OpenAI says that word error rate in its ChatGPT dictation evaluations was at least 10% lower for the top languages tested than with its previous production model ²⁰. That is a claim about the top languages it tested. It does not say how much better the tool is in a smaller language.

The gap comes from data. A 2025 review of code-switching speech recognition found that a small subset of languages receive most of the attention, and that the languages covered by datasets are very uneven ²¹. Many languages have far less recorded data than English, and the gap shows up in the scores. The guide to speech recognition accuracy by accent covers the accent side of the same problem, and the glossary entry on low-resource languages defines the term.

Accents matter too. Google says that speech recognition systems are most often trained on typical speech, so people with speech impairments or heavy accents do not get the same benefit ²². Adapting a system to accented speech can help. In a 2007 study of foreign-accented English, the best combination of pronunciation modeling and acoustic adaptation cut the word error rate by a relative 46 to 71 percent ²³.

The engine matters as much as the language. In a 2024 study of university lecture recordings, Google's average English word error rate was about 6.9 times that of Whisper large-v2 ²⁴. So if one tool is inaccurate for your speech, a different engine may do better in the same language. The glossary entry on word error rate explains how the number is calculated, and the guide to word error rate shows how to measure it.

Languages differ in speech rate, but carry similar information per second

People often ask whether they speak faster in some languages. A 2019 study of 17 languages found that they encode similar information rates, about 39 bits per second, despite large differences in speech rate and in information per syllable ²⁵. The study found large differences in speech rate and in information per syllable, but similar totals per second, so a fast-sounding language is not necessarily carrying more meaning.

That is why words per minute does not transfer cleanly between languages. Words differ in length and in the number of syllables, so the same spoken idea produces a different word count in each language. Our sources do not give a comparable words-per-minute figure for each language, so this guide does not give one. For English speaking pace, see the guide to speaking speed, and for the search answer, see the question on words per minute by language.

For dictation, the practical point is about accuracy and editing time, not raw speed. A tool that recognizes a language poorly will cost you time in corrections, however fast the speaker is.

Automatic punctuation and voice commands reach English first and other languages later

Punctuation is where languages differ most in practice. Apple says that in supported languages, Dictation on iPhone inserts commas, periods and question marks as you speak, without a spoken command ²⁶. The phrase "supported languages" is the qualifier that matters. Some formatting commands are not available for every language, such as the "No space" command on iPhone ²⁷.

Google's spoken punctuation works the same way. Gboard's voice typing might not offer punctuation phrases in every language it supports ²⁸, and Google's Docs help page warned of the same limit in 2020 ²⁹. Microsoft's Word Dictate says its preview languages may have lower accuracy or limited punctuation support ³⁰. Preview languages are a warning sign to test before you rely on them.

Punctuation rules also differ between languages, not only the word lists. If you need to say commas and full stops in another language, see the guide to dictation punctuation in other languages, and the general guide to dictation punctuation for English.

Pick a tool by checking your exact language and region, then testing it on your own speech

The built-in tools are enough for most people who dictate in a major language. Apple Dictation on a Mac or iPhone and voice typing on Windows come built in and need no setup beyond choosing the language. If you need a language those tools do not offer, or you need offline use for a language they do not cover offline, you will need a different tool. Use these steps to decide.

1. Find your exact language and region. Start with the vendor's own list, not a third-party roundup. On Windows, a language that is not installed can be added in Settings > Time & language > Language & region, as Microsoft's voice typing article describes. On a Mac, use the dictation languages in Keyboard settings, as the Mac guide to dictation languages describes.
2. Check the connection rule. Confirm whether the tool works offline for your language. Windows voice typing and Dictate in Word both need the internet ^{5 6}.
3. Check the punctuation and commands. If you dictate with spoken punctuation, test it in your language, because support can differ from one language to the next ²⁸.
4. Test on your own speech. Dictate five minutes of your own writing in the target language, then count the errors. Accuracy depends on your voice, microphone and subject, so a sample from your own work tells you more than a benchmark. The guide to common dictation mistakes covers the fixes for common errors.
5. Compare two tools if the first fails. If one tool makes many errors in your language, a second engine may do better, as the lecture study above suggests ²⁴.

If you dictate in a language that the built-in tools handle well, start with what your device already offers. If you need more, check the guide to the best dictation software and the question on which app supports the most languages. Yap, which we make, is an on-device dictation app for Mac and Windows. For its current details, see the download page.

The lists in this guide were checked on 17 October 2026, and vendors change them often. Treat each number here as dated, and check the vendor's page before you buy or rely on a language.