One model speaks all of them. Here they are, with the things actually worth knowing before you write the text.
How long will my text take?
The “chars/s” column is computed from recordings this site has already made — how many characters of that language fit into one second of speech, on average. Divide the length of your text by that number and you get the approximate length of the audio. For example 600 characters of English ≈ 46 seconds, while the same 600 characters of Japanese ≈ 90, because the script is denser.
Acronyms read letter by letter need spaces: N A S A. Contractions ("don't", "we'll") are read naturally, so keep them if you want a spoken feel.
Write in English →Write dates and phone numbers as words when the listener has to copy them down; digits are read faster but grouped by the model.
Write in Bulgarian →Long compounds are read as one word. If a compound comes out wrong, a hyphen at the seam usually fixes it: Donau-Dampfschiff.
Write in German →Keep the accents — é, è and ê are three different vowels. Punctuation decides where the liaison falls.
Write in French →Keep ñ and the accents; opening ¿ and ¡ tell the model the melody starts as a question or an exclamation.
Write in Spanish →Double consonants change the word — "nono" and "nonno" are not the same. Do not simplify them.
Write in Italian →Nasal marks (ã, õ) and the cedilla change the sound completely; ASCII substitutes will not do.
Write in Portuguese →Write ё where the word needs it — "все" and "всё" are different words and are read differently.
Write in Russian →ї, є and ґ are separate letters; substituting the Russian equivalents changes the pronunciation.
Write in Ukrainian →ł, ą, ę, ż, ź and ś are not decorations — the ASCII versions are different words.
Write in Polish →Keep the háček: ř has no substitute and a word without it will be read wrong.
Write in Czech →Long vowels (á, í, ú) and ľ carry meaning; dropping them shortens the wrong syllable.
Write in Slovak →Use ș and ț with the comma below, not the Turkish ş and ţ with the cedilla — they are different characters.
Write in Romanian →ő and ű are long vowels and distinguish words from ö and ü. Keep them exactly.
Write in Hungarian →č, ć, dž, š and ž are five distinct sounds; the plain letters read as something else.
Write in Croatian →Keep č, š and ž; without them the model reads a different word.
Write in Slovenian →“ij” behaves as one letter. Long compounds read best when written as one word, as in normal Dutch.
Write in Dutch →Keep æ, ø and å — they are letters of the alphabet, not accented versions of a, o and aa.
Write in Danish →å, ä and ö are separate letters; “a” instead of “å” is simply another word.
Write in Swedish →Double vowels and double consonants mark length and change meaning: “tuli”, “tuuli”, “tulli”.
Write in Finnish →Estonian has three degrees of length; write the doubled letters exactly as in the dictionary.
Write in Estonian →The macrons (ā, ē, ī, ū) mark long vowels and are not optional.
Write in Latvian →ą, č, ę, ė, į, š, ų, ū and ž each change the word; keep them all.
Write in Lithuanian →The tonos marks the stressed syllable. Greek without accents is read with the wrong stress.
Write in Greek →The dotted and dotless i (i / ı, İ / I) are four different letters. Copy them exactly.
Write in Turkish →Diacritics (tashkeel) resolve words that share the same consonants; add them where the meaning is ambiguous.
Write in Arabic →Write in Devanagari. Latin transliteration is read as English letters, not as Hindi.
Write in Hindi →Indonesian spelling is close to its pronunciation, so plain text usually needs no fixing.
Write in Indonesian →The tone marks are mandatory — the same syllable with a different tone is a different word.
Write in Vietnamese →For names and rare kanji, write the reading in kana. A kanji can have several readings and the model has to pick one.
Write in Japanese →Hangul is read directly. Korean has two number systems, so write numbers out when the counting word matters.
Write in Korean →Diacritics are not decoration. Every language above has at least one mark whose removal turns a word into a different word — and the model reads what it sees, not what you meant. If you paste text from somewhere that stripped the accents and tails, put them back before you press the button.
And second: the language is chosen separately from the voice. All ten voices speak all 31 languages, so pick the voice you like and change only the language.
Добави Vach на екрана
Бърз достъп като приложение
Добави Vach на екрана
Инсталирай като приложение
Добави Vach на екрана
Инсталирай като приложение