We recently released the ability to train Prototypes on WavTool. For those unfamiliar, Prototypes capture the unique sound of an instrument or voice as an AI model, allowing users to apply it to other performances. It's a revolutionary tool that unleashes creative possibilities, but with great power comes great responsibility.
Consider this scenario: I play a Chinese instrument called the ruan (阮). It’s similar to the guitar but has a mellower, almost rounder timbre. What would the guitar solo in Sweet Child O’ Mine sound like if it was played on my ruan? By applying a prototype, we can find out in seconds! Fun, right? In fact, you may be surprised to learn that the same technology can also be used to transfer the sound of one voice to another.
Check out how Prototypes can be used in the video below:
Many uses of this ability are just as fun and innocuous (imagine hearing yourself – or your little brother – performing Beyonce’s ‘Single Ladies’), but there is also a dark side.
Until now, we've provided a limited set of Prototypes for experimentation. Our ultimate goal for Prototypes, however, was to invite users to explore this astonishing new technology and create their custom models. This sparked numerous discussions within our team.
Artists are rightfully concerned about the threat generative AI could pose to their livelihoods. But many, like us, also see a path to realizing the full potential of human creativity through the very same technology.
When employed as a complete substitute for human creativity, generative AI paints a dispiriting picture of the future of art. But when used as a tool, an accelerant to human expression, generative AI becomes a source of profound optimism.
At WavTool we’re on a mission to deliver the power of AI in a form that elevates the human artist. For us, that means a fundamental commitment to AI safety. This commitment was at the core of our decision-making process when it came to Voice Model training.
We acknowledge the concerns surrounding the potential misuse of voice models, and therefore, have implemented stringent safety measures. We use safeguards to ensure that users can only create voice models of themselves or people who have consented to training a model directly, preventing impersonation and abuse. Voice models are trained only if the words spoken are at least an 80% match with a unique script generated by AI on demand, and instrument models have all vocal sounds removed as part of the training process.
By incorporating these safeguards, WavTool ensures that the power and joy of voice modeling are harnessed responsibly, eliminating the potential for abuse.
Taking safety seriously is not just a statement but a practice at WavTool. Our commitment is not only to foster creativity but to maintain a secure environment for all users. We encourage you to reach out with your feedback and concerns – our team is available on Discord or via email to address any inquiries.
Ready to try it out for yourself? Head over to the Studio!