Hear any sentence in the language you are learning
Reading a language you are learning and hearing it are different skills, and textbooks are much better at the first. Type a sentence here and a native neural voice will say it - at full speed, or slowed to half so you can hear where the words divide.
75 languages, 322 voices, many with several regional accents. Free, no credit card, and every sentence downloadable as an MP3 you can keep for listening practice.
- 75 languages
- Native accents
- Half speed
Edit the line to hear anything you like.
Four practice routines that work
- Shadowing. Generate a paragraph at 75% speed. Play it and speak along a half-second behind, copying the rhythm rather than reading ahead. Rebuild at full speed once it stops feeling rushed.
- Dictation. Make an MP3 of sentences you have not seen written down, with a pause after each. Write what you hear, then check against your own source text.
- Vocabulary with gaps. Target word, a two-second pause to recall the meaning, then the translation. Pauses are inserted in the editor, so the whole set comes out as one file.
- Minimal pairs. Put two words that you keep confusing next to each other and listen to them back to back. Hearing the contrast directly is much more effective than reading a phonetic description of it.
Which voice to choose
Accent matters more than most learners expect. If you are learning for a particular country, pick a voice from that locale rather than the language’s default.
| If you are learning… | Look for |
|---|---|
| Spanish | Spain and Latin American locales differ audibly - Mexico, Argentina, Colombia and Spain are all separate voices here. |
| Portuguese | Brazilian and European Portuguese are far apart in rhythm and vowels. Choose deliberately. |
| Arabic | Modern Standard Arabic is what most voices speak; regional locales are listed separately where they exist. |
| Chinese | Mandarin, Cantonese and Taiwanese Mandarin are distinct entries, not settings on one voice. |
| English | British, American, Australian, Irish, Indian, Nigerian and more, if you are learning towards a specific exam or region. |
Every language has its own page listing all of its voices and accents, with a box to type your own sentence and hear it before you commit - browse all 75 languages.
Closing the loop on your own speaking
The transcriber runs a speech model entirely inside your browser, so nothing you record is uploaded anywhere. Read a passage aloud, transcribe your own recording, and compare the result with what you meant to say - where the transcript disagrees with you is usually where your pronunciation is not landing.
Questions
- Are these native voices or an English accent?
- Native. Each voice is built for its own locale - a Spanish voice is a Spanish voice, not an English one reading Spanish words. Many languages have several regional accents to choose between, which matters if you are learning for a specific country.
- How slow can I make it?
- Half speed. That is slow enough to hear where one word ends and the next begins, which is the single hardest thing about a language you are new to. The player also has its own speed control, so you can drop a difficult phrase to half without regenerating.
- Can I make a listening file for my commute?
- Yes, and it is one of the better uses. Paste a vocabulary list or a set of example sentences, add pauses between items so you have time to answer in your head, and download the MP3.
- Does it handle non-Latin scripts?
- Yes - Arabic, Chinese, Japanese, Korean, Hindi, Thai, Greek, Hebrew, Russian and others are all in the catalogue. Type or paste in the language's own script and the matching voice will read it.
- Will it read a sentence with two languages in it?
- Somewhat. The multilingual voices handle a switch mid-sentence better than a single-locale voice, which will apply its own phonetics to the foreign words. For a genuinely bilingual script, the multi-speaker mode lets you cast a different voice per line, which works far better.
- Is the pronunciation reliable enough to learn from?
- For ordinary sentences, yes - these are the same neural voices used in commercial products. Be more careful with proper nouns, loanwords and unusual names, where any synthetic voice can be wrong. Treat it as a very good model rather than an infallible one.