How to change the accent in a recording
Quick answer: To change the accent in a recording, sign in, add your file, pick the accent, confirm it's your voice, change the accent and listen and download.
- Sign in. It's free. Each account gets 3 minutes of new speech a day, because every minute runs on a paid GPU.
- Add your file. Drop an audio or video file with English speech, up to 10 minutes long.
- Pick the accent. British, American, Indian, Australian or neutral English.
- Confirm it's your voice. Tick the box: it must be your own voice, or you have the speaker's permission.
- Change the accent. Whisper writes down what's said on your device. Each part (up to 20 seconds) is re-spoken in the new accent and your voice, then fitted back into its original place.
- Listen and download. Play the result on the page and download it in the same format as your file.
How does it keep my voice?
Two open-source models from MyShell, both MIT-licensed. MeloTTS speaks your words with the accent you pick. Then OpenVoice V2 measures the tone colour of your voice from a short clip (up to 15 seconds) and converts the new speech into it. The accent (rhythm, vowels, intonation) comes from the first step and the voice from the second. In our tests the converted speech matched the original speaker far more closely than the plain MeloTTS voice did, but it's not a perfect clone: expect it to sound like you on a slightly different microphone.
What is sent, and what isn't?
Your file is never uploaded. It's decoded and transcribed in your browser. For each part, the transcript text and one short clip of your voice go to our server and on to our own OpenVoice server on Hugging Face, which sends back the new audio. Nothing is stored there or on our server, and we don't log the text or audio. Mixing the result and rebuilding a video happen on your device again.
What it can't do (yet)
It re-speaks English only, so the speech in your file should be English. Background music and sounds aren't kept: you get the new voice alone. The timing follows the original sentences, so in a talking-head video the lips won't match exactly. Very fast, mumbled or overlapping speech gives a weaker transcript and so a weaker result: you can check each part's text as it goes.
Frequently asked questions
Is my audio uploaded?
Your file isn't. Short clips of your voice (up to 15 seconds) and the transcript are sent to our voice server to make the new audio, and nothing is stored. The transcription and the final mix run on your device.
Why do I need to sign in?
Every second of new speech runs on a paid GPU, so the tool is for signed-in accounts, with 3 free minutes a day. Signing in only needs an email address, with no password.
Can I change someone else's accent?
Only with their permission. You must confirm it's your own voice or that the speaker agreed. Don't use it to impersonate anyone or to mislead people. OpenVoice also adds an inaudible watermark that marks the audio as generated.
Which accents are there?
British (UK), American (US), Indian, Australian and a neutral English voice, all from MeloTTS's English model.
Does it work on videos?
Yes. Drop an MP4, MOV, WebM or MKV. The picture is copied as it is and only the sound track is replaced, so you get the same video file type back.
Is it free?
Yes, 3 minutes of new speech a day per account, reset at midnight UTC. There's no watermark you can hear and no sign-up fee.