This is different and way beyond RX. It's not just for noise removal. I just got back from NAB this morning (took the red-eye) and one of the Sony guys demoed this to me yesterday and it was absolutely amazing. It's like a cross between RX, Melodyne, and Photoshop. It has the concept of layers like Photoshop and you can select sounds like individual notes and associated them with a layer and then turn the layer's on and off. Instead of removing the noise, you can actually select what you want to keep on one or more layers and then turn off the original layer which is just the noise that you don't want without affecting the original sounds.
You can also add VST plug-ins to the layers. So imagine extracting a guitar solo or vocal from a mix onto a layer and then applying FX to that layer which only affects the guitar or vocal in the mix! It's mind boggling.
If you go to the developer's site, www.divideframe.com, you can get much more information. On their Spectral Layers comparison chart you can see the features for their various configurations, but no price. However, if you do a web search for "spectral layer price" on Yahoo and use the link to the cached page, you can see that the Pro edition is listed at $695 while the Enterprise edition is $2199.
I remember reading some technical papers years ago on "scene" detection in audio, in which sources of sound are identified and separated, based on scene analysis concepts in the visual world. I wonder if it is something like that. Different sound sources usually have overlapping spectral components, so simple spectral analysis won't do much.