Read Aloud

Text that changes language mid-sentence

Most text-to-speech tools assume a page is written in one language. Real writing often is not. A project update names a Tokyo office, a message to colleagues switches to Chinese for a phrase there is no neat English for, a Korean line arrives in the middle of an English paragraph. Hand that to an ordinary reader and it either reads everything with one accent or gives up on the parts it does not recognise.

This tool reads the text first and works out where each language starts and stops. Each stretch then goes to a voice that actually speaks it, so the Chinese is read as Chinese and the English around it is read as English.

How to use it

1. Paste your text, or press Load sample to hear four languages in three lines.
2. Watch the language map and the chips under it. They redraw as you type, so you can see where the tool thinks each language begins before you hear a word of it.
3. Press Read aloud. The marker tracks the reading and the chip being spoken is raised.
4. Click any chip to start from there instead of from the top.
5. Set a voice and a speed for each language separately in the Voices panel. Chinese read a little slower than English is often easier to follow than both at the same pace.

The hard part is the shared characters

Chinese and Japanese share a writing system. 会議室 is a perfectly good word in both, and nothing about the characters says which one is meant. Guessing wrong is not a small error: a Chinese voice reading Japanese does not sound accented, it sounds like a different sentence.

The rule this tool uses is the one a person would use. Japanese almost always mixes those characters with hiragana or katakana, so if a sentence contains either, the shared characters in it are Japanese. If the sentence has none but the paragraph around it does, it is still Japanese. Otherwise it is read as Chinese. That handles ordinary writing well and fails on a Japanese heading written entirely in kanji, which is why the setting to override it exists.

Where the voices come from

The voices belong to your browser, not to this site. That has three consequences worth knowing. The tool costs nothing to run, so it will not be taken away or put behind a sign-up. Choose a voice installed on the device and it keeps working with the network switched off. And what you hear depends on what you are using: Microsoft Edge offers neural voices that sound close to a person, Safari on a Mac offers a large local set, and a phone may have only a handful. If a language is missing entirely, the tool says so before you press play rather than falling silent halfway through.

A missing voice is worth fixing rather than working around, because it is usually a download away. Windows adds languages under Time & language, Android under the text-to-speech settings of its speech engine, and iOS and macOS under the spoken content settings in Accessibility. Add the language there, reload this page, and the new voice appears in the list.

Where your text goes

Gleekit is not in the path, whichever voice you use. There is no server behind this tool: the detection, the segmenting and the choice of voice are all code running in your browser, and none of that code sends what you pasted to us or to anybody else. We could not read your text if we wanted to, and there is no share link here that could carry it out by accident.

What your browser does next is worth knowing, because it differs by voice. A voice installed on the device is a program already on your computer, so the words never leave it and the reading works with the network switched off. A streamed voice is not installed at all: Edge's Online (Natural) voices belong to Microsoft and Chrome's Google voices belong to Google, and your browser sends your text to that company to be turned into sound, under its privacy terms rather than ours. Streamed voices are usually the best-sounding ones on a machine, so each language list puts them at the top and marks each one Streamed. If you would rather nothing left the device, pick one of the unmarked voices below them, and then switch the network off: a voice that still speaks is one running on your own computer, and that is a test no promise of ours can substitute for.

When you want a file rather than a playback

A browser will speak text but will not give you the audio, so there is no honest download button to put here. Instead the panel at the foot of the tool, below the segments, writes out a command you can run yourself. It uses edge-tts, a small open-source program, and it carries over what you set up above: the same split into languages, the same voices where they exist on Microsoft's service, the same speeds, pitch and volume. Running it leaves one MP3 on your machine, ready for a slide deck, a lesson, or a message to somebody who would rather listen than read.

None of that happens here. Gleekit never contacts Microsoft, or anybody else, on your behalf. The panel only writes a command out as text; through that route your text reaches Microsoft solely if you choose to run it, and it leaves from your own computer rather than from this page.

Good uses

• Checking that a bilingual announcement reads cleanly before you send it to the team.
• Hearing how a name, a place or a phrase sounds in a language you read but do not speak.
• Proofreading your own writing, because the ear catches repetition the eye skips over.
• Listening to a long article or a set of notes while your hands are busy with something else.
• Language practice, at a speed you choose separately for the language you are learning.