Hello all,
short question on the fluidbufnmfseed: does the coverage factor change the actual audio information preserved in the decomposed components, or just how much audio information the process considers when decomposing?
would a coverage of 0.1 or 0.9. still produce the same sum amount of audio in the resulting components or are there any spectral losses depending on the value?
thanks!
jan
Hello!
The short answer is: no change to the null-summing once you run nmf with the resulting bases as seed.
the long answer is here: BufNMFSeed reference
It is quite a complex algo to help trying to start NMF from something more intelligent than random. It takes time, so we took it out for NMF and gave it its own object.
I hope this helps!
p
Ok i see, thanks @tremblap!
i had just stumbled on the formulation:
coverage Fraction (0 to 1) of information preserved in the decomposition
so i was wondering whether the not preserved information would mean spectral loss, but i understand its only on the analysis level.