First choose a multilingual speech model and the language you intend to speak. In SuprFlow 1.0.0, model names ending .en are English-only, and the Models screen warns when one is active while a non-English language is selected. The public requirements page lists 19 supported languages. The app contains a wider internal language list, but that list alone does not prove every language is exposed, tested, or supported in the current signed release. Check your installed app and the vendor's current support statement for your specific language.
A language picker and a multilingual model are prerequisites, not an accuracy guarantee. Names borrowed from another language, acronyms, code-switching, and figures deserve a task-specific test. The model comparison guide explains how to keep the rest of the setup fixed.
Test the actual speaking pattern
Create a fictional four-sentence script: two sentences in the primary language, one containing an English product or person name, and one with an amount, date, and negation. Use the same microphone and room for each run. Record the selected language, model, app version, and formatter. Dictate the script three times, then inspect raw recognition and final inserted text separately. Count material changes rather than claiming a single word-error rate from a tiny sample.
| Check | Why it matters |
|---|---|
| Script and font | Characters and diacritics must appear correctly in the destination app. |
| Mixed-language name | The speech model may normalize or translate a name unexpectedly. |
| Amount and date | Locale-specific punctuation or number words can change meaning. |
| Negation | A missing “not” or equivalent can reverse an instruction. |
| Formatter | Cleanup or a hosted rewrite may alter the recognized language or structure. |
| Final destination | A browser form may paste or save differently from a local note. |
If a recurring fictional or approved term is wrong, try a Dictionary entry and retest. The Dictionary guides recognition; it is not a guaranteed final replacement. Do not enter real client names into a personal or unmanaged Mac just to run a test. For an official note, a fluent bilingual output still needs a reviewer who understands both languages and the subject matter.
Questions people ask
Will an English-only .en model transcribe Spanish or Hindi reliably?
The app marks .en models English-only and warns when a different language is selected. Choose a multilingual model and test the actual language in the signed release.
Does SuprFlow automatically detect a language switch mid-sentence?
Transcription uses the selected language setting. Do not assume accurate automatic code-switching; use a mixed-language script and inspect the result before relying on it.
If my language appears in the app, is it officially supported?
Not necessarily. The current public requirements page lists 19 languages, while the app's internal list may contain more entries. The installed release, model availability, and vendor support policy determine what can be promised.
Related guides
- Compare Small and Medium speech models
- Add a recurring name to the Dictionary
- Why a Dictionary correction may not work
- Test names and figures in adviser notes