Multilingual dictation on Mac without language errors
Dictating in a mix of languages on Mac without the app getting confused? A practical guide to handling code-switching and getting clean output.
You dictate a sentence in English, throw in “I have a call at 3” or “the deadline is Friday,” and suddenly everything goes sideways: the app switches languages, writes things phonetically, or turns “call” into something unrecognizable. Code-switching is part of daily life for anyone working in tech, marketing, or international business. The problem isn’t you — it’s that most dictation tools force a single language per session. Here’s how to dictate in multiple languages on Mac without everything falling apart.
Why dictation gets the language wrong
macOS’s built-in dictation and most third-party apps work with a fixed language setting. You choose “English” in preferences, and everything is interpreted as English — including any foreign words you say. The result: a technical term from another language gets fed through an English model that “hears” the sounds and transcribes them with approximate English spelling.
Code-switching (mixing languages within the same sentence) breaks this model. Three common scenarios cause trouble:
- A single borrowed word: “I pushed the commit to the main branch” — one word from another language inside an otherwise single-language sentence.
- A full sentence switch: you dictate an email to a French colleague, then switch back to English for your notes.
- Industry jargon: sprint, backlog, churn, pipeline… words borrowed from other languages that have become everyday work vocabulary.
The solution: automatic detection + LLM rewriting
This is exactly what sets modern dictation apart from raw transcription. With Speech Flow, the approach is different: you hold Ctrl, you speak, and two things happen.
First, the transcription engine handles multiple languages without any manual configuration. Then — and this is the key — an LLM reviews the transcript and cleans it up: it strips filler words (“um,” “like,” “so”), adds proper punctuation, and most importantly preserves foreign words as-is rather than forcing them through the wrong language model. “Bonjour” stays “Bonjour,” “schadenfreude” stays “schadenfreude.”
In practice: how it handles code-switching
The LLM has the context of the entire sentence. When you say “the meeting has been pushed to tomorrow,” it understands that borrowed words are intentional and writes them correctly, without attempting to transliterate them phonetically. That’s the difference between a corrector that guesses word by word and a model that reads the whole sentence before writing anything.
| What you dictate | Native dictation (fixed language) | Speech Flow (LLM + context) |
|---|---|---|
| “I have a réunion at 3” | “I have a reunion at 3” | “I have a réunion at 3.” |
| “the deadline is vendredi” | “the deadline is Von Freddy” | “The deadline is vendredi.” |
| A fully foreign sentence | words mangled to English sounds | clean text in the right language |
How to dictate in multiple languages effectively
A few simple habits for reliable results:
- Speak naturally. Don’t exaggerate the accent on borrowed words. Say them as you normally would — the model keeps up.
- Don’t switch settings between languages. The whole point is that you never touch the preferences. You just dictate.
- Use full sentences. The LLM relies on context: a complete sentence is handled far better than a single word dropped in isolation.
- Proofread rare proper nouns. No tool is perfect on an obscure brand name or a made-up acronym. A quick glance is all it takes.
Speech Flow natively supports multiple languages (EN, FR, ES, IT). You can dictate an email in English, your notes in French, and a message in Spanish back to back, without reconfiguring anything.
What about privacy?
Fair question, since an LLM is involved. Speech Flow works on a BYOK (Bring Your Own Key) model: you bring your own API key (OpenAI, Gemini, or Groq). No audio is stored, and processing goes through your account with the provider of your choice. That’s more transparent than a black-box service that holds on to your recordings.
If you’d rather not manage a key yourself, an all-inclusive plan provides the keys for €10/month or €70/year.
FAQ
Do I need to select a language before dictating?
No. That’s the whole point: detection is automatic, and the LLM preserves foreign words inside a sentence without forcing them into the dominant language.
Is code-switching actually handled well, or just “tolerated”?
Actually handled: the model reads the entire sentence, so it knows a foreign word dropped into English is intentional and spells it correctly, rather than phonetically mangling it.
How is this better than macOS built-in dictation?
macOS dictation locks you into one language and doesn’t rewrite the text at all. No filler-word cleanup, no smart punctuation, no meaningful code-switching support. For a full comparison of dictation apps, see our Speech Flow vs Wispr Flow page.
In summary
If you juggle languages all day, a dictation tool locked to a single language costs you time in constant corrections. A transcription + LLM-rewriting approach handles code-switching instead of breaking on it. Speech Flow is native macOS (Apple Silicon), lightweight (~50 MB), and costs €69 lifetime on the BYOK plan. If mixing languages is part of your daily workflow, take a look at the options — no magic promises, just dictation that stops mangling every foreign word you say.