Speak however it arrives
You do not have to choose a language before you open your mouth, and you do not have to stay in it. A sentence that starts in one language and finishes in another is one sentence, not two failures.
Languages
Speech recognises over 120 languages, and writes back what you said rather than a translation of it — unless a translation is exactly what you asked for.

Two separate settings
Most dictation ties those together: pick a language and it will only hear that one. Speech keeps them apart, which is the difference that matters if you do not live your whole life in one language.
You do not have to choose a language before you open your mouth, and you do not have to stay in it. A sentence that starts in one language and finishes in another is one sentence, not two failures.
Leave the output alone and Speech writes your words down. It is not trying to improve your language or move you into English.
Choose an output language and every line lands in that one instead, in its own script. That is the setting to reach for when you are writing to somebody else’s inbox.
Add the people, products and places you mention to your dictionary once. Proper nouns are where every dictation tool fails first, and it is the fastest fix available to you.
The “um”, the false start and the sentence you abandoned halfway are removed. The meaning, the numbers and the names are not touched.
Privacy Mode skips saved dictation audio and transcripts, including local history, on the free plan as much as a paid one.
If you live in two languages
It started with Hindi and English, because the people who built it write that way all day and nothing on the market could handle it. What came out of that is a recogniser that treats a mixed sentence as ordinary rather than as an error — which is just as true of Spanish and English, or Arabic and French, as it is of the case it was built for.
The case it was built forNot on a sentence written to be tested. Open the thing you were about to type anyway and say it the way you would have said it.
Download SpeechOver 120. Recognition and written output are separate lists, though: what Speech can hear is much longer than the list of languages it will translate INTO.
Only if you ask. Leave the output setting alone and it writes the language you spoke. Language models are not perfectly obedient and an occasional translated line is the failure we watch most closely, so tell us when you see one and include the sentence.
That is the case Speech was built around. It is treated as one sentence rather than as a recognition error.
Add them to your dictionary and they are spelled your way from then on. It is the single change that makes the biggest difference to everyday accuracy.
Yes. Recognition and cleanup run online on every platform.
Because it is made in India and most of its users are there. Prices switch to dollars for you automatically; the control on the pricing page overrides it either way.
Free to begin, unlimited on Android, and no card to get started.