Free Italian Speech-to-Text

Italian has around 65 million native speakers. It is the official language of Italy and San Marino, the working language of Vatican City, one of the four national languages of Switzerland — where some 8 % of the population gives Italian as their main language — and a recognised minority language in Istria and along the Slovenian coast. It is also a young standard: when Italy unified in 1861, only a small minority used Italian in daily life, and most people spoke a local Romance language instead. Italian spread through schooling, then radio and television, which is why almost every Italian recording carries a regional layer underneath the standard.

The first thing Italian demands of a transcript is consonant length. Doubled consonants are not decoration — they change the word. Casa is a house and cassa is a crate; nono is ninth and nonno is a grandfather; caro is dear and carro is a cart; sete is thirst and sette is seven; papa is the Pope, papà is your father and pappa is baby food. The only difference the ear gets is how long the consonant is held, so a system that hears Italian has to measure duration and then commit it to spelling. What makes this genuinely awkward is that the same lengthening also happens between words: in central and southern Italy "a casa" is pronounced with a long c, [akˈkaːsa], but it is still written with one. The sound doubles, the spelling must not.

Then there are the marks above and between the letters. A grave accent is often the whole difference in meaning: è is "is" and e is "and", dà is "gives" and da is "from", sì is "yes" and si is the reflexive pronoun, né is "nor" and ne is a particle. Modern Italian writes perché, poiché and né with an acute accent, not a grave. Apostrophes mark elision rather than possession — l'amico, dell'anno, d'accordo — and the rules are narrow enough to catch people out: un'amica takes an apostrophe because it is feminine, un amico does not, un po' takes one because it is a truncation of poco, and qual è takes none at all. On top of that sits the silent h, which exists in Italian for almost no reason except to keep homophones apart: ha and a, hanno and anno, ho and o are identical in the ear and can only be settled by grammar.

ConvertSpeech handles all of this automatically. You upload the file and the engine writes out the spoken Italian as punctuated, searchable text you can paste into a document, a subtitle file or a translation tool. There is nothing to install, it runs in your browser, and your first minutes are free with no account required.

This is a real 45-second stretch of an Italian audiobook recording, put through our AI engine exactly the way your own file would go through it. Nothing was tidied up afterwards — what you read below is what came back, slips included. Press play and follow along.

Worth noticing how much of this is spelling rather than hearing. Ve l'ho già detto fuses lo and ho into l'ho and keeps the silent h that separates the verb from the vowel o. L'altro and d'economia carry their apostrophes. Perché arrives with the modern acute accent, even though the printed edition this recording was read from sets it the old way, perchè, with a grave. The doubled consonants land where Italian requires them — settimana, bottega, avvertire, accadesse, sufficiente — including carro in non mettiamo il carro avanti ai buoi, which is a different word from caro. Dapprima and pianterreno are written as the single words modern Italian spells them, where the old edition splits them into da prima and pian terreno. The names Spinello Spinelli and Fiordalisa survive intact, and the engine even closes the line of dialogue with a caporale, », where the printed page uses a dash.

It is not flawless, and both slips sit exactly where you would expect them: at the two cut edges, where the engine has only half a phrase to work with. The clip opens mid-sentence and the book's per filo e per segno comes back as A filo per segno. It ends mid-sentence too, and the original's question — la ragione sufficiente? I filosofi… — is resolved as la ragione sufficiente: il filosofo! Neither was corrected here. Beyond that, remember what this recording is: one trained voice reading carefully into a quiet microphone, roughly the difficulty of a lecture or a dictated note. A market-square interview with three people talking over each other will not come back this clean, and nothing automatic will. Recording quality remains the single biggest factor in the result.

Standard Italian grew out of the Tuscan of Florence — the language of Dante, Petrarch and Boccaccio, later polished by Manzoni — and it is what broadcast, academic and professional recordings use. ConvertSpeech is tuned for it and handles it reliably across speakers, ages and speaking speeds.

The north–south divide is the one you will actually hear. Roughly along the La Spezia–Rimini line, the doubling of consonants across word boundaries stops: a Roman or a Neapolitan says "a casa" with a long c, a Milanese or a Torinese does not. Northern speakers voice the s between vowels, so casa sounds closer to "caza", while central and southern speakers keep it sharp. Tuscans soften c between vowels almost to an h — la casa as "la hasa", the gorgia toscana. And southern speakers routinely use the passato remoto in everyday narration where a northerner reaches for the passato prossimo, so the same story comes back as andai, dissi, feci in Naples and sono andato, ho detto, ho fatto in Milan. All of it is correct Italian and all of it transcribes well; it just means the tense and register of your transcript will reflect where your speaker is from.

Swiss Italian is its own case. It is spoken in Ticino and the southern valleys of Graubünden and is fully standard Italian in structure, so it transcribes as reliably as the Italian of Italy. What differs is a thin layer of vocabulary — helvetisms, mostly calques from the German and French of Swiss administration. Azione is a supermarket special offer rather than an action, riservazione stands in for prenotazione, comandare is used where Italy says ordinare, and a mobile phone can still be a natel. These come through as written, and they are not mistakes: they are what the speaker said.

Dialect is the real limit, and it is worth being precise about why. Neapolitan, Sicilian and Venetian are not accents of Italian — they are separate Romance languages that developed in parallel from Latin, and each holds its own ISO 639-3 code (nap, scn, vec). Sardinian and Friulian go further still and are protected as minority languages under Italian law 482/1999. A speaker using standard Italian with a Neapolitan or Sicilian accent transcribes well. A speaker switching into actual Neapolitan or Sicilian is speaking a different language, and no Italian-language system — ours included — will reproduce it faithfully; expect an approximation rendered in standard Italian spelling. For strongly dialectal material, clean audio and speakers staying in standard Italian give by far the best result.