Supply two different recordings
Voice Changer needs a source performance and a reference voice. The source contains the words and delivery to convert; the reference supplies the vocal characteristics. Load or record both, preview them, run conversion and listen before downloading.
The input limit is 60 MB per file and clips shorter than half a second are rejected. A clean, single-speaker reference is more useful than a busy music mix. Use your own voice or a reference you have permission to use, and disclose synthetic alterations where appropriate.
This page converts recordings after they are supplied. It is not a live-call voice changer, a language translator or a reliable anonymity tool. Conversion can alter pronunciation and timing, so do not treat the result as an unchanged evidentiary recording.
How reference conversion works
Both recordings are prepared as mono audio at 22,050 Hz. The OpenVoice2 model uses the source performance and reference voice through voice-encoding and conversion stages to produce the result.
Preparation, voice encoding, conversion and export are separate stages. Progress milestones indicate completed stages, not a promise that progress advances uniformly in time.
WAV output is written as 32-bit float audio. That representation does not undo model changes or prove that the output exactly matches the reference speaker. Long source clips can still require considerable memory and computation even when compressed input files are small.
Compare conversion with model packs and effects
| Website | Method exposed to the user | Important distinction |
|---|---|---|
| Mediatavio | OpenVoice2 conversion with source and permitted reference recordings | Reference-conditioned recorded-audio conversion; no live-call integration |
| LALAL.AI Voice Changer | Upload audio/video, choose a voice model and adjust documented settings | Voice packs and hosted processing; its separate Voice Cloner creates a voice model |
| VoiceChanger.io | Audio or microphone input with pitch, filtering, echo and other effects | Signal effects are not equivalent to reference-conditioned neural voice conversion |
This table paraphrases the linked official pages. No shared source/reference pair was tested, and no identity-similarity or speed ranking is claimed. Review rights, consent and each service's data terms before uploading a voice.
Evaluate words as well as timbre
Keep the same source, reference and listening level. Check intelligibility, missing syllables, breaths, pitch jumps and unwanted background sounds. A voice that sounds more like the reference but changes important words may be the worse result for narration.
For a straightforward musical pitch adjustment, use Pitch Shifter instead. To generate a new script, use Text to Speech. For music stem preparation, first inspect the output of Vocal Remover.