Dictating Names Across Languages Without Misspellings
Dictation is fast until it meets a name it has never heard. Suddenly Nguyen becomes "win," Siobhan becomes "shove on," and Wojciech turns into three unrelated words. Here is how to dictate names across languages and keep every one of them spelled correctly on your Mac.
Key takeaways
- Names get mangled because they are rare in the model's training data, not because you spoke badly.
- A custom dictionary is the single biggest fix: teach the app the spelling once and it holds.
- On-device AI cleanup applies the correct name consistently across a whole document.
- Running on-device keeps sensitive names, like clients or patients, entirely on your Mac.
Why dictation mangles foreign names
Speech-to-text does not spell. It predicts. A model listens to a stream of sounds and outputs the most statistically likely sequence of words it was trained on. English names appear constantly in that training data, so "James" and "Sarah" are easy. A Polish, Vietnamese or Yoruba name appears far less often, so the model reaches for a common English word that sounds close instead. That is why the error is almost always a real, correctly spelled word, just the wrong one.
Accents and diacritics add a second layer. The sounds in Zoë, Jokūbas or Björk carry information that plain English phonetics throw away, so even a good transcription can drop the mark. If you want the deeper background, the mechanics are the same ones described in the general overview of speech recognition, and modern local engines like Whisper handle multilingual audio far better than older systems, though rare names still trip them.
None of this is your pronunciation failing. It is a probability engine guessing, and the good news is that you can change the odds. To see which tongues your current setup even attempts, check what languages Mac dictation supports before you blame yourself.
The five-step method that actually holds
You do not need to fight the model word by word. Set it up once with these steps and the fixes stick across every document you dictate.
Collect your recurring names
List the people, places and brands you dictate often: colleagues, clients, product names, cities. These are the words worth teaching the app once.
Add them to a custom dictionary
Enter each name exactly as it should appear, diacritics included. BlaBlaType's custom dictionary is built for names and jargon, so the spelling is locked in.
Speak the name a touch slower
Keep your normal pace for the sentence, then give the name a clear, unrushed beat. Cleaner audio gives the model a cleaner target to match.
Let on-device AI clean the pass
AI cleanup fixes punctuation and applies your dictionary consistently, so if a name slips through once it is corrected everywhere in the same edit.
Add a custom prompt for edge cases
For tricky mixed-language notes, a short custom AI prompt like "keep all proper nouns as spelled in the dictionary" reinforces the behavior.
What the custom dictionary changes in practice
The custom dictionary is doing the heavy lifting, so it is worth understanding the shift. Without it, every dictation is a fresh guess and you spend time proofreading names you have typed a hundred times. With it, the model has an anchor: when the incoming sound is close to an entry, it commits to your spelling instead of inventing a plausible English word. That is the difference between correcting Wojciech in every email and never seeing the error again.
| Situation | No custom dictionary | With custom dictionary |
|---|---|---|
| Common English name | Usually correct | Correct |
| Foreign name (Nguyen, Siobhan) | Often mangled | Held to your spelling |
| Names with accents (Zoë, José) | Marks dropped | Preserved |
| Brand or product name | Split or misspelled | Consistent |
| Same note, second mention | May differ from first | Matches every time |
This is also why non-native speakers get so much value from it: the dictionary compensates for the model's blind spots regardless of your accent. We go deeper on that in our roundup of the best dictation apps for non-native speakers, and on where the technology stands in how accurate voice-to-text is in 2026.
Keep every name spelled right
Custom dictionary, on-device AI cleanup and 90+ languages, all running locally on your Mac. No card needed for the trial.
Download for macOSPrivacy matters when the names are real people
Names are personal data. If you dictate client lists, patient notes or the roster of a deal under NDA, where those names are processed is not a footnote. Cloud dictation uploads your audio to a server to transcribe it, which means those names leave your machine. On-device dictation does not: the speech recognition and the AI cleanup both run on your Mac, so the names, and the audio that contains them, never travel. If this is a concern for your work, read our full take on whether Mac dictation is private. You can also compare plans on the pricing page to see what stays local.
Mini glossary
- Custom dictionary
- A user-defined list of exact spellings that overrides the model's guesses, so names and jargon appear the way you intend.
- Diacritic
- An accent mark such as the é in José or the ø in Søren that changes a letter's sound and must be preserved to spell a name correctly.
- On-device processing
- Transcription and AI cleanup that run entirely on your own Mac, so audio and text are never uploaded to a server.
- AI cleanup
- A post-transcription pass that fixes punctuation and grammar and applies your dictionary consistently across the whole text.
- Proper noun
- The name of a specific person, place or brand, the category of word dictation is most likely to misspell without help.
Putting it together
Dictating names across languages stops being a chore the moment you stop treating each name as a one-off. Teach the app your recurring names, give them a clear beat when you speak, and let on-device AI apply the fix everywhere. The names you dictate every week get spelled right the first time, and the private ones stay on your Mac. Most people speak around three to four times faster than they type, so once names stop breaking that flow, dictation finally pays off for multilingual work.
Frequently asked questions
Why does dictation misspell foreign names?
Speech-to-text models predict the most likely spelling from the sounds they hear. Names from other languages are rare in their training data, so the model guesses a common word that sounds similar. A custom dictionary tells the model the exact spelling you want.
How do I stop Mac dictation from changing names?
Add the names to a custom dictionary so the app locks in the spelling, speak the name a little slower and more clearly, and let on-device AI cleanup apply the correction consistently across the whole document.
Can I dictate names in several languages at once?
Yes. On Mac, BlaBlaType supports 90+ languages and a custom dictionary, so you can dictate a mostly English note that contains German, Polish or Japanese names and keep each one spelled correctly.
Does dictating names send my data to the cloud?
It depends on the app. BlaBlaType runs speech recognition and AI cleanup 100% on-device, so the names you dictate, including private client or patient names, never leave your Mac.
What is the fastest way to fix a name the model keeps getting wrong?
Add the exact spelling to your custom dictionary once. After that the model maps the sound to your spelling every time, so you stop retyping the same name and stop hunting for typos later.