Here’s an interesting presentation of an Adobe research prototype for finding several instances ofthe same sound in an audio recording.
This is good! I like it for the 2 angles:
- on the one hand, you have a black-box, optimised, focused process (similar-ish instances of sound in duration, none overlapping, etc) with an interesting interface (the curve is good and in line with other discussions we have had in the lab and on this forum)
- on the other hand, you can hear world leading people having ‘research code’ results - artefacts of splitting, algo not working all the time, etc
This is inspiring! Thanks for sharing
Okay seriously, who in the crowd is payed enough to scream and clap when they see concatenative synthesis happen at a tech conference. Does anyone have a connection here that they can hook me up with?
You are so mean! What is impressive is the click = (segment properly) (make a stencil) (find the same with a curve) (segment that curve)
but yes, a bit over-enthusiastic, especially for the annoying actor-cum-mc …
Yes and I assume most of the benefit of whatever technique their using is derived from having $$$ to throw at training deep learning models.
but yes, this is nice to see. A lot of this is unpolished as you say, “dirty underwear” and you can see all the remnants of their research code in the output. Python and the big one that popped out is that this is a matplotlib window with some GUI to interface with the more complex code. I reckon there are probably ways that this kind of slicing and selection based on similiarty could be done quite well with the tools to help.