[Eridrus]

This already exists for Google Assistant devices, but is not what the article is talking about.

The article implies that it will fuse voice data from multiple devices which both only heard part of the query into a single query, which is a lot easier to write down in a patent than to actually make work.