Text to Speech

This text to speech tool reads your text out loud and, with its AI voice, saves the speech as a WAV or MP3 file. It is for students, writers, people who would rather listen than read, and anyone who needs a quick voice-over for a video or slide deck.

You get two engines on one page. Instant uses the voices already on your phone or computer, starts speaking at once and covers every language your system has, Bengali included on many Android phones. Kokoro AI is an open voice model that runs inside your browser and sounds far more natural in English; it downloads once and your text never leaves your device.

There is no sign-up, no character counter that stops you at 2,000 letters and no watermark in the audio.

Use Instant to just listen, for text that isn't English, or on a slow connection. It starts in under a second because the voice is part of your system. Newer Windows, macOS and Android voices sound good; older ones sound robotic.

Use Kokoro when you need a file or a more natural English voice. Kokoro-82M is an open model with 82 million parameters under the Apache 2.0 license, so you can use the audio commercially. It needs a one-time download and real computing power.

• Instant: any language your device has, no download, word highlighting, cannot be saved as a file.
• Kokoro on a graphics chip (WebGPU): about 359 MB once, then faster than real time after the first sentence, even on built-in graphics.
• Kokoro on the processor: about 113 MB once, but slow, around 15 to 25 seconds of work per second of speech. Fine for a paragraph, not for a chapter.

Device voices are played by the operating system straight to your speakers. The Web Speech API that browsers give to web pages has a speak command but no way to capture the sound it makes, so no website can save it as a file without asking you to record your whole screen or tab.

Kokoro is different. The page itself computes every audio sample, so it can hand you a WAV or MP3 file.

Here, American and British English. The model has voices for seven more languages, but each needs its own pronunciation engine, and the usual free one (eSpeak NG) is GPL-licensed. English pronunciation comes from the misaki dictionaries made for Kokoro: about 180,000 words per accent.

Bengali is not supported by Kokoro at all. For Bangla, use the Instant engine with a Bengali voice. Many Android phones have one through Google's speech services, and Windows can add one with the Bengali language pack. The tool tells you if it finds one.

WAV is uncompressed 16-bit audio at 24,000 samples per second, the rate Kokoro produces. One minute is about 2.9 MB. Pick it for editing, since every cut and fade starts from full quality.

MP3 at 96 kbps is about 0.7 MB a minute and plays everywhere. The LAME encoder (LGPL) loads as a separate, unmodified module only when you ask for an MP3.