Admittedly, I’m not certain how best to use this function. That said, I have selected guitar transients and attempted to select similar and it’s taking a LONG time.
Granted, I am using a long music selection, so taking a long time might be reasonable. What seems unusual is that both CPU and GPU usage while in this state are running in the 2-5% utilization range!
Once completed, I can’t say that the results are what I expected, but the main point is understanding why this seems to take forever while simultaneously showing extremely low CPU/GPU utilization.
I haven’t used SpectraLayers enough to have an intuitive grasp of what to expect, results-wise, but I do not expect to see extremely low results while also seeing low CPU/GPU utilization.
Select Similar I also find incredibly slow, yet I use a 2018 computer build, so CPU only at this time for me.
I have found Select Similar to return reasonable results for finding things like cricket chirps, but man, once the human using SLP can easily recognize such a pattern manually, it is a pretty simple manual task. Cicada on the other hand, is much more randomized due to isolation of a single cricket is fairly easy to capture and cicada is always a massive crowd, so a lot of randomization.
Can I ask what sound you are looking to Select Similar for?
Getting the parameters set just right for select similar is a task in and of itself, especially considering how long one pass of select similar might take.
I’ve tried using it for a couple of things. My first try was to capture the transient portion of guitar. Frankly, I wasn’t sure if it would work at all and the results were less than stellar, as it tended to locate tonal portions as well, and it seemed to capture other transients.
As you indicated, the problem with experimenting with how this works is massively restricted by the extremely slow performance. I certainly can understand that this might be a CPU intensive process! What seems unusual is that it is not consuming CPU/GPU resources, leaving me baffled about what it’s doing to take so long.
To expand on your comments, I understand that some creativity may be required to translate visual identification into programmable parameters. Having said that, it does seem that a focus on how to identify via program, things that are readily identifiable visually would be a huge upgrade to functionality.
Now, I’m not saying this is necessarily a trivial task! However, an expanded SELECT that would allow some detection of frequencies with certain harmonics, detection of repeated patterns, etc. might ultimately be very helpful.
I’ve noticed in very dense mixes with layers of instruments, that playing around with the min and max levels can make individual instruments easier to see visually and ultimately making selecting those parts easier. So now, how do I then identify and program those features into SELECT to assist in identifying things that the normal automation miss?