English text to speech

How to actually improve with this, not just listen to it

Playing a paragraph back once and nodding is not practice. Here is the loop that works, and it takes about four minutes.

1. Read it aloud yourself first. Before you press play. This matters more than anything else on the page: if you listen first, you will copy what you just heard and learn nothing about what you would have done. Say it your way, then check.

2. Play one sentence, slowed down. Use the speed control at 0.7 or 0.75. At full speed English word endings vanish into the next word and you will not hear whether the -ed was there. Slow speech is not natural speech, but it is diagnostic speech.

3. Listen for three things specifically. Which syllable is loudest in each long word; whether the sentence goes up or down at the end; and where the speaker does not pause. Learners almost always pause too often, at commas, which is where English speakers frequently do not.

4. Say it again with the audio. At the same time, out loud, over the top of the voice. This feels ridiculous and it is the single most effective thing on this list — it forces your timing onto someone else's rhythm instead of your own.

5. Then speed it back up to 1.0 and check you can still keep up. If you can, that sentence is done.

Why hearing your own writing is different from listening to a podcast

There is no shortage of English listening material, and this tool is not competing with any of it. Listening to real speakers is how you learn to understand English. It is not, on its own, how you learn to produce it — because when you listen to a native speaker you are hearing sentences you did not write, containing words you might never choose.

The gap most intermediate learners are stuck in is different: they can read and write far better than they can say. They construct a good sentence on paper and then cannot get it out of their mouth, because they have never heard it, and because English spelling does not tell you what a word sounds like. Thought, through and thorough are not a joke; they are a real problem you cannot solve by reading.

Putting your own sentence through a voice closes that specific gap. The sentence is already yours — you chose the words, you know what it means, and there is no comprehension work in the way. All that is left is the sound, which is the thing you were actually stuck on.

This is also the fastest way to find out that a word you have used for years is a word you cannot say. Almost everybody has several. Our pronunciation pages cover the sounds themselves, with minimal pairs and the British IPA for 5,500 words.

What a synthetic voice is honestly good for

The voice reading your text is your browser's speech synthesiser, not a recording of a person. It is very good at some things and cannot do others at all, and knowing which is which is the difference between a useful tool and a misleading one.

Trust it for: word stress, the shape of a word ending, the rise and fall of a statement or a question, and the speed real speech actually runs at — which is faster than most learners expect. On all four it is reliable enough to correct yourself against.

Do not trust it for: emphasis and emotion, which it flattens; regional accent, which it has one arbitrary version of; unusual names and rare words, which it will guess at from spelling and sometimes get wrong; and — most importantly — any judgement about your own speech, because it cannot hear you.

It is a model of an accent, not a teacher. Use it to check the shape of a sentence, then use a human being to check yourself. If you have nobody to practise with, our guide to practising English speaking alone is about exactly that problem.

Why it reads one sentence at a time

The reader splits your text into sentences before it speaks a word, and gives each one its own play button. That is a deliberate design decision with two reasons behind it.

The first is pedagogical. Repetition is what fixes pronunciation, and repetition means going back to this sentence twenty times without listening to the four before it again. A single play button on a whole paragraph makes that impossible, which is why almost nobody practises properly with one.

The second is technical, and it is the sort of thing worth knowing if you use text-to-speech elsewhere. Browser speech engines truncate very long utterances — pass a whole essay in one go and it will simply stop somewhere in the middle, usually without telling you. Splitting at sentence boundaries first means every chunk is comfortably short, so nothing is silently cut off.

A worked example

This is the example text loaded by the button above the box, and how the reader breaks it up. Five sentences, five buttons, each replayable on its own.

1. Last summer I decided to write to my old teacher.
2. I had not spoken to her for eleven years, and I was not sure she would remember me.
3. She wrote back in three days.
4. Her letter was two pages long and she remembered everything, including the essay I had been so proud of.
5. I have read it four times since then.

Notice that the split is on sentence-ending punctuation, not on commas or line breaks, and that abbreviations such as Dr. do not start a new sentence. A sentence longer than about four hundred characters is capped rather than truncated mid-word — if yours is longer than that, it is worth breaking up for a reader anyway.

That paragraph is also worth a second look for what it does to a learner's mouth. It contains wrote and written, thought and three, and an -ed ending said three different ways — decided /ɪd/, remembered /d/, proud of with the linking that makes it one word. Slow it to 0.7 and every one of those is audible.

Frequently asked questions