https://www.youtube.com/watch?v=u_rYgikYDG0&t=922s&ab_channel=WhiteNoiseStudio
I’m going to share a tip (some insider knowledge):
Because I know how some of these datasets/libraries/dictionaries are built and work, you can actually get dramatically different results (IN A NON-DESTRUCTIVE FASHION) by adding a little noise.
In SpectraLayers there is a process (in the generate section) called “Noise”. If you process any audio and add some white noise to it (meaning you select the generate process and select the Noise process) and select mix mode and add about -27.0db level noise to it and run it through any of those algorithms, it dramatically(in most cases) changes the separation process and you get dramatically different results. If you phase flip the original layer against the noise layer, then you can use those results in a non-destructive fashion. Again, your results may vary but it is something that people should be aware of when using these algorithms and comparing different results.
2 Likes