About voicechange.io

voicechange.io is a free AI voice changer that runs in your browser. You can record or upload your voice and hear it in someone else's, type text and have it spoken in that voice, turn a song into a cover sung by a different singer. There is no account, nothing to install and no watermark on what you download. For live use in Discord and games, and to clone your own voice, there is a free desktop app.

What you can do here

  • Change your voice. Record up to 5 minutes in the browser or upload an mp3, wav, m4a, ogg or webm file (up to 15 MB). The words, timing and emotion stay yours; the tone of voice becomes the one you picked.
  • Text to speech. Type a script and hear it read in any of the 140 voices in the catalog, up to 1,000 characters at a time.
  • AI song covers. Upload a song (up to 8 minutes). The vocals are split from the music, converted to the new voice and mixed back over the original instrumental.
  • Your own voice. Clone your voice with the free VoiceChange desktop app. Voices you trained on the website before stay in the "Your voice" tab.

How it works

Each voice in the catalog is a Retrieval-based Voice Conversion (RVC) model: a neural network trained on recordings of one speaker or singer. When you send audio, the system first measures two things in your recording: what is being said (the phonetic content) and the pitch contour of your voice. The voice model then rebuilds the audio from those two signals using the timbre it learned from its training data. That is why the result keeps your delivery, timing and accent but sounds like a different person.

Pitch is handled separately. Every voice has a measured centre pitch, and "auto pitch" shifts your recording toward it before conversion, so a deep voice going into a high character voice (or the other way round) does not come out strained. You can override it with the pitch slider. The pitch guide goes into detail.

Text to speech uses an open-source speech synthesizer (Kokoro) to read your text in a neutral voice, which is then converted to the target voice with the same RVC step. Song covers use a source-separation model to pull the lead vocal out of the mix first; only the vocal is converted, and the instrumental is added back at the end.

All conversion runs on our own GPU server. Your audio is never sent to a third-party AI service.

Where the voices come from

Most catalog voices are community-trained RVC models that hobbyists have shared publicly. New voices are mainly added from visitor requests: when you ask for a voice that is missing, we look for a suitable public model, download it, test it by converting a fixed sample sentence, measure its pitch range and only publish it if the test passes. Every voice page has that sample clip so you can hear the model before using it. If a voice sounds wrong, tell us and it gets fixed, replaced with a better model, or removed.

These are fan-made imitations. voicechange.io is not affiliated with, sponsored or endorsed by any of the artists, actors, characters or rights holders whose voices the models imitate.

Responsible use

Voice changers are mostly used for memes, parody, fan content, music experiments and accessibility, and that is what this site is for. Please do not use it to deceive people into thinking a real person said something, to impersonate someone for fraud, or to harass anyone. Only train a model on your own voice or on a voice you have permission to use. The full rules are in the terms of use.

Your audio and privacy

You do not need an account. Your browser gets a random ID so it can show your recent results and your trained voices; we do not ask for your name or email. Uploaded audio is used only to produce your result. The details, including the cookies used by advertising, are in the privacy policy.

How the site is funded

The GPU time behind every conversion costs money. voicechange.io stays free for everyone by showing ads.

Who runs it

voicechange.io is built and run by an independent developer. The site is updated continuously: new voices are added most weeks, and bugs reported through the contact address get fixed. If a voice sounds wrong, a page has a mistake, or you represent someone whose voice is in the catalog and want it removed, email [email protected] or see the contact page.