search

neuroscout / neuroscout

NeuroScout web app and API
https://neuroscout.org
BSD 3-Clause "New" or "Revised" License
15 stars 12 forks source link

Extract TF audio features #784

Open PeerHerholz opened 4 years ago

PeerHerholz commented 4 years ago

Ahoi hoi gang,

within our discussion wrt auditory features and things (e.g. #702 and #596), I wanted to propose the inclusion of the cochleagram feature (standard version here and tensorflow version here) which could act as a new, adjustable low-level feature. This would also allow us to integrate DNN layer based features from this pre-trained network. This could be utilized for both music and speech (looking in your direction @rbroc). For references please see here.

adelavega commented 4 years ago

hey @PeerHerholz circling back to this. If we're going to include this, let's first include it into pliers.

Any advice on "standard" vs tf version? I assume only the latter would work for the pre-trained network?

Let's get the ball rolling on some of these features! Want to meet sometime soon?

PeerHerholz commented 4 years ago

Yeah, sure thing, thx for getting back to this. I didn't compare the standard and tf version directly, I used the standard version a while back and it also worked with the pre-trained network. Next week is so far pretty free for me.

adelavega commented 4 years ago

Hey, next week would be better actually, I'm moving this week. @rbroc, want to chat about this too?

PeerHerholz commented 4 years ago

Sure thing, just let me know!

rbroc commented 4 years ago

Would be great! Next week works for me too (best from Weds on, but earlier on can also work).