Skip to content

Language Support

FloWords supports 99+ languages through the Whisper models. This guide covers language configuration, auto-detection, and tips for multilingual transcription.


These languages have the highest accuracy and are well-tested:

LanguageCodeNotes
EnglishenBest support, use .en models
SpanishesExcellent accuracy
FrenchfrVery good support
GermandeVery good support
ItalianitVery good support
PortugueseptIncluding Brazilian
DutchnlGood support
RussianruGood support
JapanesejaGood support
ChinesezhMandarin, good support
KoreankoGood support

Solid accuracy for most use cases:

LanguageCode
Arabicar
Czechcs
Danishda
Finnishfi
Greekel
Hebrewhe
Hindihi
Hungarianhu
Indonesianid
Norwegianno
Polishpl
Romanianro
Swedishsv
Thaith
Turkishtr
Ukrainianuk
Vietnamesevi

Available but may have lower accuracy:

Click to see all 99+ supported languages

Afrikaans, Albanian, Amharic, Armenian, Azerbaijani, Bashkir, Basque, Belarusian, Bengali, Bosnian, Breton, Bulgarian, Burmese, Cantonese, Catalan, Croatian, Estonian, Faroese, Galician, Georgian, Gujarati, Haitian, Hausa, Hawaiian, Icelandic, Javanese, Kannada, Kazakh, Khmer, Lao, Latin, Latvian, Lingala, Lithuanian, Luxembourgish, Macedonian, Malagasy, Malay, Malayalam, Maltese, Maori, Marathi, Mongolian, Nepali, Occitan, Pashto, Persian, Punjabi, Sanskrit, Serbian, Shona, Sindhi, Sinhala, Slovak, Slovenian, Somali, Sundanese, Swahili, Tagalog, Tajik, Tamil, Tatar, Telugu, Tibetan, Turkmen, Urdu, Uzbek, Welsh, Yiddish, Yoruba…


Set a specific language for consistent results:

  1. Open Settings (⌘,)
  2. Go to Model tab
  3. Find Language dropdown
  4. Select your language
  5. Close Settings

Let Whisper detect the language automatically:

  1. Go to Settings > Model
  2. Set Language to Auto or Detect
  3. Whisper will identify the language per recording

For English-only users, FloWords offers .en models:

ModelStandardEnglish-Only
Tinytinytiny.en
Basebasebase.en
Smallsmallsmall.en
Mediummediummedium.en
  • Faster: No language detection overhead
  • More Accurate: Optimized specifically for English
  • Smaller: Slightly smaller file sizes

If your recording contains multiple languages:

  1. Set language to Auto-Detect
  2. Use a Large model for best accuracy
  3. Review results for language switches
  4. Manually correct if needed
  • Rapid language switching may confuse the model
  • Heavily mixed content (code-switching) is challenging
  • Some language pairs work better than others
  • Try to separate content by language when possible
  • Use explicit language setting when one language dominates
  • Larger models handle mixed content better

Whisper handles various English accents:

  • American English
  • British English
  • Australian English
  • Indian English
  • South African English

For best results with strong accents, use larger models.

Supports regional Spanish:

  • Castilian (Spain)
  • Latin American Spanish
  • Mexican Spanish
  • Mandarin (best support)
  • Cantonese (basic support)
  • European Portuguese
  • Brazilian Portuguese

  • Use .en models for best performance
  • Small.en offers excellent balance
  • Technical terms: add to Custom Dictionary
  • Set language explicitly (es)
  • Medium model recommended
  • Regional vocabulary may need dictionary additions
  • Larger models significantly improve accuracy
  • Consider cloud providers for complex content
  • Technical terms often in English
  • Mandarin has better support than Cantonese
  • Large model recommended
  • May need post-processing for tone marks
  • Right-to-left output handled correctly
  • Large model recommended for dialectal Arabic
  • Modern Standard Arabic best supported

Symptoms: Text appears in wrong language or script

Solutions:

  • Set language explicitly instead of auto-detect
  • Check audio doesn’t have mixed languages
  • Use a larger model

Symptoms: Many errors, garbled output

Solutions:

  • Use Medium or Large model
  • Check audio quality
  • Reduce background noise
  • Speak more clearly
  • Try a cloud provider

Symptoms: Characters display as boxes or question marks

Solutions:

  • Ensure your Mac has appropriate fonts
  • Check system language settings
  • Copy text to a Unicode-capable application

Symptoms: Strong accents poorly transcribed

Solutions:

  • Use larger models
  • Speak slightly slower
  • Consider cloud providers with accent support
  • Add common mispronunciations to dictionary

Use CaseRecommendation
Single language, EnglishUse .en model
Single language, otherSet language explicitly
Occasional mixingAuto-detect + large model
Professional multilingualConsider cloud providers
Accuracy NeedModelLanguage Setting
Quick notesTiny/BaseExplicit
Standard useSmallExplicit
ProfessionalMedium/LargeExplicit
Mixed contentLargeAuto-detect

Whisper continues to improve, and FloWords will support new language capabilities as they become available. Updates include:

  • Improved accuracy for existing languages
  • Better handling of accents and dialects
  • Enhanced mixed-language support