Very interesting quick geeky nmf~ explanations

Feel free to let me know at what slide you get lost… and which ideas excite you too!

2 Likes

Disclaimer: as soon as it gets math-heavy, I’m not more interested. But your mileage may vary. I find some of the promises in the first slides to be quite exciting :wink:

this is great. i’m now down a maths rabbit hole. also found this video useful (for me, maybe not anyone else):

not that it has given me thoughts about this project, but if there was a real-time version it would massively help with the research project i’m doing with music and cochlear implant users. oh look:

2 Likes

This is such a great explanation, thanks for sharing ! It goes on the dark side from 1:10 but in a visual way so that is good. I wonder what @jamesbradbury and @rodrigo.constanzo would do of it. I presume too mathy? I start loosing interests at 2:00 :wink:

1 Like

Indeed get’s really mathy really quick, but an interesting watch nonetheless.

The biggest take away from my rabbit hole last night was that rank selection is a whole research area, and approaches can be low-rank or solving all ranks in a range and then selections being made based on ‘quality’ measures (by code not humans). Of course we’re most likely going to be doing it by ear. I suppose the more I try it with different types of material I might start to intuit something about the relationship between material and rank choice, and several suggestions for this approach have already been made on the forum. I do wonder if there is a quick and dirty pre-analysis that could be done which might suggest a suitable rank range. Just random thoughts pre-coffee…

1 Like

Also, I reckon part 2 of this project is going to be fun…

Also, I haven’t read the “There are 111 replies with an estimated read time of 40 minutes” thread yet so this probably answers much of what I’m asking here…

2 Likes

this project is amazing! @weefuzzy and @groma check the paper! Cross synth with nmf, my dream :wink: