Voice to Text in 25+ Languages on Mac

Hapi transcribes over 25 languages locally on your Mac, no cloud processing, using NVIDIA's Parakeet multilingual model. Before you dig into a specific language, two honest caveats:

1. We've directly tested English and Spanish ourselves. The other languages below inherit Parakeet's published multilingual coverage — real, but not something we've personally stress-tested against native speakers yet. If a language misbehaves for you, that's useful signal, not an edge case we'd rather not hear about.
2. There's no language picker to configure. Parakeet auto-detects per utterance. The "Transcription language" setting in Hapi only flags a mismatch after the fact — it doesn't force output into a language.

Language coverage table

Language	Script / notes	What makes it harder to transcribe	Guide
Spanish	Latin	Regional accents (Castilian vs. Latin American), fast pace, frequent subjunctive	Full guide
Dutch	Latin	Guttural sounds, compound words, Dutch/Flemish variants	Full guide
Chinese (Mandarin)	Hanzi	Tonal (four tones), no word boundaries in writing, homophones	Full guide
Hindi	Devanagari	Retroflex consonants, frequent code-switching with English (Hinglish)	Full guide
Russian	Cyrillic	Palatalized consonants, stress-dependent vowel reduction	Full guide
Hebrew	Hebrew, RTL	No written vowels, formal/colloquial register gap	Full guide
Arabic	Arabic, RTL	Dialectal variation (MSA vs. regional), no short vowels in writing	Full guide
German	Latin	Compound nouns, umlauts, Sie/du formality distinction	Full guide
Turkish	Latin	Agglutinative (long suffixed words), vowel harmony	Full guide
Japanese	Hiragana/katakana/kanji	Three writing systems, context-dependent word boundaries	Full guide
Portuguese	Latin	Brazilian vs. European variants, nasal diphthongs	Full guide
Indonesian	Latin	Affixed word forms, formal/informal register	Full guide
Polish	Latin	Seven grammatical cases, consonant clusters	Full guide
Swedish	Latin	Pitch accent, the sje-sound	Full guide
Thai	Thai script	Five tones, no spaces between words	Full guide
Korean	Hangul	Agglutinative grammar, honorific speech levels	Full guide
Vietnamese	Latin + diacritics	Six tones, monosyllabic word structure	Full guide
Romanian	Latin + diacritics	Special characters (ă, â, î, ș, ț), Latin-Slavic vocabulary mix	See below
Czech	Latin + diacritics	Consonant-heavy words, háčky/čárky diacritics	See below
Ukrainian	Cyrillic	Distinct from Russian, soft sign (ь), unique phonemes	See below
Catalan	Latin	Vowel reduction, distinct from Spanish, ç and l·l	See below
Italian	Latin	Geminate consonants, regional dialects	See below
Greek	Greek script	Stress-based accent system, consonant clusters	See below
French	Latin	Liaison between words, nasal vowels, silent consonants	See below

(Chinese, Hindi, Arabic, and Hebrew ship as full standalone guides too — linked above.)

Less-common languages, covered here

These seven don't have enough dedicated search traffic yet to justify a separate page each — the content lives here instead of splitting it across seven thin pages.

Catalan

10+ million speakers. Catalan is a distinct Romance language, not a Spanish dialect — Hapi treats it as its own language rather than folding it into Spanish output. Main transcription challenges: vowel reduction in unstressed syllables and characters like ç and l·l that don't exist in Spanish.

Italian

85+ million speakers. Geminate (doubled) consonants carry real meaning in Italian — pena ("pain") vs. penna ("pen") — which is exactly the kind of distinction that's hard for any ASR model to catch from audio alone. Regional dialects (Sicilian, Neapolitan, Venetian) diverge enough from standard Italian to add further variance.

Greek

13+ million speakers. Greek's own script plus a stress-based accent system (an accented vowel changes which syllable is emphasized, not just pronunciation) makes it a genuinely distinct case from the Latin-script languages above.

Romanian

24+ million speakers. Romanian is the outlier Romance language — Latin grammar with a heavy layer of Slavic vocabulary from centuries of regional contact, plus diacritics (ă, â, î, ș, ț) that don't map cleanly onto any other language in this list.

Czech

10+ million speakers. Consonant-heavy word formation (words with no vowels at all, like scvrnkls, are grammatically valid) and the háček/čárka diacritic system are the two things that make Czech transcription meaningfully harder than its Slavic neighbors.

Ukrainian

45+ million speakers. Cyrillic, but a distinct alphabet from Russian's — Ukrainian has letters Russian doesn't (ґ, є, ї, і) and drops others Russian has. Treating Ukrainian as "close enough to Russian" is a common mistake that shows up in transcription quality, not just translation.

French

300+ million speakers. Written French keeps letters that spoken French drops entirely — liaison rules mean the same word sounds different depending on what follows it, and nasal vowels (the "on" in bon) have no equivalent in English phonetics. Silent consonants at word endings are the norm, not the exception.

Try it yourself

Free tier, no account, no subscription. Download Hapi and speak — Parakeet auto-detects the language.