Multilingual Voice Typing on iPhone

Three settings people mix up

Almost every multilingual voice typing problem on iPhone comes from confusing three settings with different jobs.

• Keyboard language: the layout, autocorrect, and suggestions you type with. In Auri these are downloadable packs, more than 80 of them, under Auri > Settings > KEYBOARD > Keyboard Languages.
• Recognition language: what the speech engine listens for. In Auri it is the Language row under Settings > KEYBOARD > Keyboard Dictation > Recognition, with 58 languages plus Auto-Detect.
• Translation target: the language you convert finished text into. Auri's Translate tool covers 31 languages.

Choosing a keyboard language does not change the recognition language. You can download the German pack, swipe to it, and dictate, and the engine still listens in whatever the Language row says. Set the two separately every time you switch.

Check recognition first whenever output looks like phonetic gibberish: speak Spanish while the engine expects English and you get English words that sound vaguely like what you said. Check the keyboard language when autocorrect fights you word by word.

Set up the two or three languages you actually use

Keep the list short: every extra pack adds a step to the switch cycle.

1. Open Auri > Settings > KEYBOARD > Keyboard Languages. Packs you have appear under On This Device; everything else sits under Available to Download.
2. Download the two or three languages you message in.
3. Switch by swiping across the Spacebar. The app's own instruction: “Swipe quickly across the Spacebar without pressing to switch through your enabled languages.”
4. Set the recognition language to match what you are about to speak: Settings > KEYBOARD > Keyboard Dictation > Recognition > Language.
5. Dictate two sentences per language into Notes and check the output.

The step people skip is the fourth. Swiping the Spacebar switches the keyboard pack only; if the Language row still says English, your Spanish take comes back as English-sounding nonsense.

On-device transcription is English only

Auri Keyboard's engine list has one on-device option, and it transcribes English only. Pick another language while On-Device is selected and the app shows an “English Required” alert and points you back to the online engines. Every other language transcribes in the cloud.

So there is no offline multilingual dictation on iPhone: dictate in Spanish, German, or Hindi and your audio goes to a cloud engine, which stops without a network. What leaves the phone in each mode is in the AI keyboard privacy guide.

What breaks when you code-switch

Bilingual speech is normal and recognition engines are bad at it. Each take is transcribed as one language, so the engine resolves the other language into whatever sounds closest.

One language per take fixes most of this. Finish the thought, stop, switch both settings, speak again. Two short takes beat one mixed take you have to rebuild.

Write directly or write then translate

Both routes work; pick per message by how fluent you are in speech versus writing in that language.

In Auri the translate route is one tool away: tap the Auri Writing key on the toolbar, choose Translate, and pick the target from 31 languages. It translates your selection if you have one, otherwise the whole draft, and replaces the text in place.

You said (recognition set to English): "can we move the call to Thursday afternoon"

Translate to Spanish: "¿Podemos mover la llamada al jueves por la tarde?"

Translation is strongest on plain, literal sentences. Speak in short declaratives, avoid idioms from your first language, and name the thing instead of gesturing at it. Read the result before sending even when you are not confident in the language: names, numbers, and dates are checkable without fluency.

Words that fail in every language

Some items fail in every language because they were never common words in any training data.

• Your own name and hometown, especially with diacritics.
• Colleague and client names that cross language boundaries.
• Brand and product names spelled unlike how they sound.
• Street addresses and postal codes.
• Field-specific terms borrowed from another language.

Auri's fix is Custom Vocabulary, on the same Recognition screen: turn on Use Custom Vocabulary, tap Add Words, and enter the names, product terms, and uncommon words the engine should recognize. The Ultrafast, Supernova, Legacy, and Azura engines use the list; Wizper and On-Device do not. The custom words guide covers what to add first.

For the rest, keep a note with the correct spellings of the twenty terms you use most and paste from it. Check accented characters every time: a name that loses its accent looks careless to the person reading it.

A daily routine for mixed-language chats

This routine fits in the seconds before you start talking.

1. Decide the language of the message before you tap the field.
2. Swipe the Spacebar to that language's pack, then check the recognition Language row matches. One does not follow the other.
3. Speak one take, one language, one point.
4. Scan names, numbers, and accents.
5. Translate only if you drafted in the other language on purpose.

When a chat moves between languages every few messages, batch by language. Answer the Spanish thread, then the English thread, and you change settings twice rather than eight times.

After a take in a language you are still learning, one Fix Grammar pass is enough; more passes drift the meaning.

Limits worth knowing

Knowing these limits prevents an embarrassing message.

• Formality is easy to lose: machine output can pick the wrong register for your reader.
• Idioms translate literally and land strangely.
• A model may produce phrasing from the wrong region.
• Legal, medical, and contractual wording needs a fluent human reader.
• Offline, nothing multilingual works: the only engine that runs without the cloud is On-Device, and it is English only.

For high-stakes messages in a language you do not write confidently, draft by voice for speed and have a fluent colleague read it before it goes out.

Auri Keyboard's default engine, Ultrafast, streams your audio to the cloud while you speak, whatever the language. If a conversation covers something you would rather keep off a server, dictate it in English on the On-Device engine or type it. Does voice typing send audio to the cloud covers the same question for every setup.

FAQ