You have a short video and a script, and you would rather not record yourself — no quiet room, no microphone, or simply no wish to hear your own voice on it. What you want is a narration track you can drop onto the timeline.
The short path
- Open the text to speech tool. The first visit downloads the model, about 64 MB, once — after that it is cached and works offline.
- Paste the line you want spoken. Up to 500 characters per generation.
- Pick one of the 19 voices, or clone one from a short sample of your own.
- Generate, listen, adjust the wording if the phrasing lands badly, and download.
Nothing is uploaded: the model runs in your browser, so neither your script nor your voice sample leaves the machine. For a client's script under embargo, that is the difference between a tool you can use and one you cannot.
What 500 characters actually is
Roughly 80 words, or about 30 seconds of speech. That covers a short-form video in one or two passes, and it is the wrong shape for a twenty-minute documentary.
For anything longer, generate one line per shot rather than one file for the whole script. That is not a workaround for the limit — it is how you would want the audio anyway. Per-shot files can be moved, retimed and replaced independently, and re-recording one sentence does not mean regenerating the narration.
Cloning a voice
The tool can take a short recording and speak new text in that voice. What makes the difference:
- A clean sample. The clone copies what it hears, including room noise and hiss. If your sample is noisy, run it through the noise reducer first — that is one of the more useful ways to pair the two tools.
- Ordinary speech, not performance. A neutral read gives a more usable voice than a dramatic one, because the clone reproduces the delivery as well as the timbre.
- A few seconds is enough, and more is not obviously better. Try a short sample first.
A recording of a real person can carry rights of publicity or personality on top of copyright, and in several jurisdictions cloning a recognisable voice without consent is unlawful regardless of who owns the recording. Your own voice, or one of the 19 built-in voices, avoids the question entirely. See Terms, section 2.
Getting a better read
- Punctuate for the ear. A comma buys a short pause and a full stop a longer one. Splitting a long sentence in two usually fixes pacing faster than any setting.
- Spell awkward words the way they sound. Names, acronyms and units are where synthesis most often goes wrong, and rewriting is quicker than fighting it.
- Generate a line twice. Output varies slightly between runs; the second take is sometimes the one you want.
- Download as WAV if it is going into an editor, and leave the compression to the final export.
Where this is the wrong tool
- Long-form narration. An audiobook or a long documentary means hundreds of generations. A server-based service with a document-length input is the right shape for that.
- Phones and older machines. The engine needs
SharedArrayBufferand a capable processor. Some mobile browsers and embedded webviews do not provide it. - Fine prosody control. There is no SSML, no phoneme editor and no per-word pitch. You steer the read by rewriting the text.
- A metered connection. The first load is a real 64 MB.
Create subtitles from the narration
Open AI Speech to Text with your narration recording to create an editable transcript and export SRT or VTT subtitles for your video editor. Each file can be up to 5 minutes long.
Frequently asked questions
How do I add a voiceover to a video without recording myself?
Type the narration into a text-to-speech tool, choose a voice, and download the audio for your timeline. A browser-based tool does this without an account and without uploading the script — the model runs on your own device.
How much text can I generate at once?
Up to 500 characters per generation, which is roughly 80 words or about 30 seconds of speech. For a longer script, generate one line per shot; that is easier to edit than a single long file in any case.
Can I use my own voice?
Yes. The tool clones a voice from a short sample, and your sample never leaves your device. Use a clean, plainly-spoken recording — the clone reproduces room noise and delivery along with the timbre.
Can I use the generated voiceover commercially?
We place no restriction on the audio you generate and claim no rights over it. The constraint is the voice: cloning a real person without consent can infringe rights of publicity regardless of copyright. The built-in voices and your own voice avoid that.
Why is the first load so large?
Because the model has to reach your browser before anything can be generated — about 64 MB, once, then cached. That download is what buys you a tool that needs no account, no quota and no upload.