Your voice makes a conversation feel human. A familiar rhythm, a particular tone, the way you say someone’s name: together, those details make you recognizable. When you publish a podcast, join a game or go live, you share more than words. You share a little of your vocal identity.
Sharing a recording should be a choice
Creators want people to hear their work. Gamers want to participate without giving every stranger the same access to their personal identity. Teams need clear, natural conversations. None of those goals requires giving someone permission to imitate your voice.
That is the question behind VocalShield: can we keep speech useful for communication while making its identity cues less useful to unauthorized AI voice-cloning systems? It is a different objective from simply removing background noise or choosing a character voice.
A protective layer, built on research
VocalShield builds on our latest scientific research into protective perturbations. These are carefully designed changes to an audio signal, aimed at interfering with the identity information a voice-cloning model tries to reuse. The goal is to preserve understandable speech while disrupting that reuse.
In Audiomni, we are developing this as a layer in the outgoing voice workflow. You should be able to see when it is active, understand which microphone carries the processed signal and quickly mute that output. Voice changing can shape how you sound; VocalShield has the separate job of targeting cloning risk.
The complete audio path matters
A local preview is only one part of a real call. Enhancement, voice effects, compression and the meeting app’s own processing can all change the audio that reaches another person. We treat those combinations as part of the protection problem, alongside speech quality and end-to-end delay.
Protection therefore needs context: which model, which settings and which audio path were evaluated? An enabled indicator should describe the processing state. It should not be read as a guarantee that every future cloning system will fail.
Privacy also starts on your device
Audiomni’s desktop direction is real-time, on-device audio processing. Your voice should not need a cloud round trip to be enhanced, reshaped or protected. Our browser audio tools follow the same principle: audio processing runs on your device, with an initial model download for tools that need one.
For now, make deliberate choices about the recordings you publish and check the input device selected in each app. As Audiomni develops, our aim is to make those decisions easier to see and easier to control, in the same place as the rest of your voice setup.
You should be able to share what you say while keeping a say in how your voice is used.
Explore VocalShield in Audiomni ↗