Recognition is not support
When a dictation app says it supports 99 languages, it almost always means one thing: the speech model it ships was trained on audio in 99 languages and will produce words when it hears them. That is recognition. It says nothing about whether the app punctuates your language correctly, applies its capitalisation and number rules, formats dates the way your country writes them, or cleans up filler words the way it does in English.
This is why the language counts on marketing pages are so misleading. An app can genuinely recognise Icelandic and still be a bad Icelandic dictation tool, because everything after recognition — the punctuation model, the style pass, the custom-dictionary handling — was built and tested for English first and a handful of large languages second.
The practical rule: treat a vendor's language count as the size of the recognised set, and ask separately about the polished set. The polished set is usually between five and thirty languages, and vendors rarely publish it. Our per-language scores in the ranking exist precisely to measure the difference.
What to check for your language specifically
Punctuation behaviour. Does the app produce your language's punctuation — inverted question marks in Spanish, guillemets in French, correct comma placement in German subordinate clauses — or does it apply English rules to foreign words? This is the single biggest quality gap between languages in the same app.
Number, date and currency formatting. 'Twenty euros fifty' should become €20.50 with your locale's conventions. Most apps only do this for English.
Mixed-language speech. If you code-switch — a Spanish sentence with an English product name in the middle — check whether the app handles it or forces you to pick one language per session. Automatic language detection is still mostly a cloud-app feature; local Whisper apps usually make you set the language manually.
Custom vocabulary. Names, jargon and dialect words matter more in languages with less training data, because the base model is weaker there. An app with a good custom-dictionary feature can close much of the gap.
The state of play, honestly summarised
English is a solved problem in every app we rank. Spanish, French, German, Portuguese, Italian and Dutch are well served by the top tier. The large Asian languages — Mandarin, Japanese, Korean — are well recognised but formatting quality varies the most between apps. The long tail of the Whisper list, from Amharic to Welsh, is recognised everywhere and polished almost nowhere.
If you dictate in a smaller language, the honest recommendation is to test two or three apps with your own voice and your own vocabulary for a week each. Recognition quality for your accent and your subject matter will differ more than any feature list suggests, and every serious app in the ranking has a free tier or trial that makes this cheap to do.
References
- OpenAI — Whisper model card and language list — the 99-language list every whisper.cpp app inherits
- NVIDIA — Parakeet TDT 0.6B V3 — the 25-language European model changing the local-dictation tradeoff
Frequently asked questions
Is a 99-language app better than a 10-language app?
Not necessarily. The 99-language apps share the same underlying Whisper model, so the count tells you what is recognised, not what is handled well. A 10-language app that punctuates and formats all ten properly will beat a 99-language app for those ten.
Which dictation app is best for Spanish?
See our dedicated Spanish guide. The short version: the proprietary cloud apps currently produce the cleanest Spanish punctuation, while the local Whisper apps offer full offline recognition with rougher formatting.