Bart

@bartblaze @janl Certainty is fairly high; that's _A LOT_ of audio to send back and centrally process vs. doing it locally

@dakami @bartblaze they might analyse locally and just send the results back home.

@janl @bartblaze generally you send back a set of "features" (independent unique fuzzy metrics). The server matches to known content.

@dakami @janl I wonder how big their set of known content is though - maybe @alexstamos has some extra information regarding this feature?

@bartblaze @janl @alexstamos it's probably pretty large; content id databases have been maintained for years and years

@bartblaze @janl @alexstamos but it's for copywritten content, not arbitrary speech

@dakami @janl @alexstamos Interesting, which kind of copywritten content would that comprise exactly?

@bartblaze @janl @alexstamos basically anything that would make it to the Pirate Bay