AudioShake uses A.I. to recognize different components in a piece of audio--for example, the drums in a rock song, or the dialogue in a film  We then isolate that stem so you can use it for new purposes--localization, sync licensing, remastering, re-mixes, and more.

No. AudioShake specializes in a different field known as source separation. The technology is quite different–most notably, it doesn’t generate any new content. Rather, it’s taking what already exists and separating it into its different components (“stems”). 
‍
Here’s a practical way to think about it: If you want to generate a music track in the style of Elvis or clone Tom Cruise’s voice, your AI would need to have a concept of what Elvis’s music or Tom Cruise’s voice sounds like-–meaning, you would need to have training data of their music or voices, or soundalikes. 
‍
In contrast, a source separation model like AudioShake doesn’t need to train on Elvis’s music in order to separate an Elvis track. It’s sufficient to train on data–for example, licensed production library data–that teaches the model the general qualities of guitars, voices, drums, etc. so it can separate any audio regardless of the specific artist or speaker. 

No. This is a risk with generative AI models, not sound separation models. Our models separate what’s already in the audio. So if there is no duck quacking in your jazz trio, there’s not going to be a duck quacking in the outputted stems.

No. Unless stated otherwise contractually–for example, in the event we are building a custom model for a rightsholder–we do not train on any input or output to or from our platform. Instead, we license training data from a wide range of providers.

Yes! All our separations are via API, and many are also available on-device. Our documentation site is here.

Yes. We’ve been through multiple AI Reviews across some of the world’s largest companies–and have passed all of them.