Long text to speech

What happens when you paste a long text

The text is split into short parts of a sentence or two each, and the parts are generated in order. You do not have to wait for the end: on most recent computers playback starts after about a second while later parts are still being prepared. On slower devices the tool waits until enough audio is ready to play without stuttering, and tells you how long that wait will be.

When everything is done, a parts list appears. Click any part to play from that point, which is the easiest way to jump back to a paragraph you missed without scrubbing through a long file.

How long it takes

On a typical recent laptop, generation runs several times faster than real time, so a text that takes an hour to listen to is ready well before the hour is up. Most phones and older computers can be slower than real time. The tool still works on them; the full file simply takes longer to finish than it takes to hear, so for very long pieces a laptop is the more comfortable choice.

The first time you use the tool there is a one-time download of the voice engine, and on some computers the first generation also needs up to a minute of one-time preparation. After that it starts almost instantly. Keep the page open while a long text is generating, since the work happens on your device rather than on a server.

Fixing one sentence without redoing an hour of audio

With long text, something almost always needs a second pass: a name read the wrong way, a heading that runs into the next line, a typo you only notice when you hear it. You can edit a part's text directly in the parts list and regenerate just that part, or change the sentence in the main box and press Generate again. Either way, unchanged parts are reused instantly and only the edited ones are made again.

For names that come out wrong, spell them the way they sound, for example 'Win' for 'Nguyen'. Numbers, dates, percentages and common units are already read in their spoken form, so those rarely need touching.

Choosing a voice and speed for long listening

Listening fatigue is real over long stretches. A voice that sounds fine in a two-minute clip can become tiring after forty minutes. Use the short listen button in the voice menu to shortlist a few voices, then generate one full paragraph of your own text with your favorite before committing to the whole piece. Heart is the default and the most natural; Nicole is softer and slower, which some people find easier to follow in dense material.

Speed goes from 0.7× to 1.5×. For material you already know, such as reviewing your own report, 1.1× or 1.25× saves real time. For unfamiliar or technical text, stay at 1× or 0.9×; you lose more by rewinding than you gain by going faster.

Splitting very long works into files

There is no limit on length, but one enormous audio file is awkward to use. For a book, a course reader or a long report, generate one chapter or section at a time and download each as its own file, numbered so they sort in order. Batch processing of many files is not available, so this is a manual step, but it leaves you with chapter-sized files that are easy to resume and share. The text to audiobook guide goes further into this workflow, and if your source is a PDF, start with PDF to speech.

Your draft is kept in the browser between visits on the same device, so if you work through a long text in stages, whatever you left in the box will still be there when you come back.

Frequently asked questions