Use TypingVoi for Multilingual Writing

This guide is for people who write in more than one language and want TypingVoi to stay predictable. It shows how to separate profiles by language, when to use Auto instead of an explicit spoken language, how to choose models for multilingual work, and how to avoid cleanup settings that accidentally rewrite the wrong way.

1. Start with one profile per regular spoken language

If you dictate in multiple languages every week, create separate profiles for the languages you actually speak out loud.

That setup is usually better than keeping one mixed profile because each language can keep its own:

• spoken language setting
• speech-to-text model choice
• vocabulary list for names and local terms
• cleanup behavior
• output behavior for the app or workflow

A practical starter setup looks like this:

• English Writing
• Spanish Writing
• Vietnamese Notes
• Bilingual Drafting only if you truly switch languages inside the same session

Use Create Profiles for Different Apps and Workflows first if you have not separated your workflows yet.

2. Use explicit language when the profile has one stable spoken language

An explicit spoken language is the safest default for multilingual writing. If one profile exists for one language, select that language directly instead of leaving the profile on Auto.

Use an explicit language when:

• you usually dictate full sentences in one language per session
• you want more stable model recommendations
• you want fewer recognition mistakes on grammar, punctuation, and common words
• you are building a vocabulary list for names or terms in that language

Examples:

• English Writing profile set to English
• German Client Notes profile set to German
• Japanese Drafts profile set to Japanese

This keeps the profile easier to tune because TypingVoi is not guessing the language every time you start.

3. Use Auto only when language switching is part of the real job

Auto is useful, but it should solve a real workflow problem instead of acting as a default guess.

Use Auto when:

• you regularly switch languages between sessions and do not want to swap profiles every time
• you dictate short segments in different languages during the same part of the day
• you are capturing mixed-language notes where a fixed language causes obvious friction

Avoid Auto when:

• one profile is clearly dedicated to one language
• you are troubleshooting accuracy and need fewer variables
• you want the best possible model recommendation for one language

In practice, Auto is best for convenience. An explicit language is usually better for consistency.

4. Pick models based on the kind of multilingual work you do

Model choice matters more in multilingual writing because the wrong tradeoff creates cleanup work in two places: recognition errors and language confusion.

Use this comparison:

• choose a stronger model when you are writing long drafts, client-facing text, or anything expensive to correct by hand
• choose a faster model when you need quick replies and can tolerate a little more manual editing
• choose a model that clearly supports the language you speak most in that profile instead of forcing one general-purpose setup across all languages

If you use separate profiles by language, each profile can make a different tradeoff. For example:

• a fast English profile for chat replies
• a stronger French profile for formal writing
• a lightweight mixed-language profile only for quick capture

Do not assume one model will be the best choice for every language you use. Start with the top recommendation for each profile, then compare only if the output is too slow or too error-prone.

Use Choose Languages, Models, and Custom Vocabulary for the profile-level tuning workflow, and Download and Manage Models if you need to install or compare additional models.

5. Treat vocabulary as language-specific, not global

Vocabulary works best when it stays close to one language and one workflow. In multilingual setups, that usually means keeping separate vocabulary lists inside separate profiles.

Add vocabulary terms for:

• people’s names
• company names
• product names
• place names
• technical terms
• recurring borrowed words the model keeps writing incorrectly

Examples of good profile-specific vocabulary use:

• an English profile with English product names and teammate names
• a Spanish profile with Spanish client names and local place names
• a Vietnamese profile with local names that a general multilingual model often misses

Avoid building one giant mixed-language vocabulary list unless the profile itself is intentionally mixed-language. If a name or term only matters in one language workflow, keep it there.

6. Be careful with names and shared terms across languages

Some multilingual workflows reuse the same names across languages, but not always with the same pronunciation or spelling expectations.

Use these rules:

• keep the canonical written form you want in the profile where it is used most
• add aliases only for recurring wrong forms
• test the final transcript, not just live preview
• split the term into different profiles if one spelling should win in one language but not another

This is especially important for:

• person names pronounced differently across languages
• brand names spoken with local accents
• English technical terms inside non-English dictation
• acronyms that should stay unchanged regardless of surrounding language

7. Keep cleanup prompts conservative for multilingual content

Cleanup AI can help readability, but multilingual content needs extra caution. A cleanup model can over-normalize text, translate unintentionally, or rewrite terms that should remain exactly as spoken.

Use cleanup carefully when:

• one transcript includes two languages
• the transcript contains code, commands, brand names, or product terms
• you need the output to remain in the same language as the source

Safer cleanup habits for multilingual writing:

• start with a conservative preset
• keep temperature closer to accuracy than creativity
• tell the model to preserve the original language of each passage
• tell it not to translate names, commands, or technical terms
• leave cleanup off while you are still testing speech-to-text accuracy

If a profile regularly mixes languages in the same transcript, test cleanup with real examples before you rely on it. In many multilingual workflows, the best order is:

1. get transcription accuracy right first
2. add vocabulary for repeated names and terms
3. enable cleanup only after the raw transcript is stable

Use Clean Up Transcripts With AI if you need to tune the model, prompt, or failure behavior in more detail.

8. A practical multilingual setup that works for most people

If you want a stable setup without overthinking it, use this pattern:

1. Create one profile for each language you use often.
2. Set an explicit spoken language in each of those profiles.
3. Start with the top recommended model for each language.
4. Add a short vocabulary list with names and terms that matter repeatedly.
5. Leave cleanup off until the raw transcripts look reliable.
6. Create one separate Auto profile only if your real workflow switches languages often.

That gives you a predictable default and still leaves room for one flexible profile when your work is less structured.

Related guides

• Create Profiles for Different Apps and Workflows
• Choose Languages, Models, and Custom Vocabulary
• Download and Manage Models
• Clean Up Transcripts With AI

Best next step: build one explicit-language profile for the language you use most, test it without cleanup first, and add only the vocabulary terms that the model actually misses.