Dear all,
Now that the new 0.9.1 release is up, and all computations seems to be working just fine. We would really like to see what the software can do in the long run ! So I am creating this sticky post so that we can all share our best sounding examples and workflows.
We saw that experience with the software bring greater results, so we believe that everybody can benefit from sharing their different experiences.
We’ll hope to be amazed very soon 
All the best,
Philippe
Hello,
The .9.1 version is much more a pleasure to work with, merci! The Play function works well until there are just too many layers to render. Is it necessary that it try to do the mix live like that? Wouldn’t it be better to actually render the mix for auditioning? There are significant differences between the mix previewed by Play and the mix rendered by Export as audio file.
I’m attaching some examples for you to hear what I’m referring to. The target file is a single Kenong stroke. The Analysis settings are PartialsMeanAmplitude and SpectralCentroidMean, Harmonic filtering is on, algorithm is Genetic. Everything else is “out of the box”, no filtering. I used Soundflower–>MaxMSP-sfplay~ to record Orchids Play-auditioning the solutions. The Exported files are what Orchids produces based on the same solutions. The differences between the Play and Export versions of the same solution are quite dramatic. There are significant differences in the relative loudness of the instruments and also some noticeable differences in entry timing.
I also wonder about the very loud presence of the violin tremolo articulation in “kenong orchestrated - Play 2.aif”. It’s less noticeable in the Export version, but still seems uncharacteristically loud in relation to anything heard in the target file. I’ve noticed this out-of-character, stentorian violin presence several times before when working with other target files.
I can see now that being able to play the target and solutions at the same time to compare them is useful. It would be nice to be able to synchronize the playback so they “lay” against each other accurately. I found myself playing a kind of game trying to get the solution lined up with the looping target playback. Or is there a better way?
But, overall, the results are promising. Working with relatively simple target sounds gives a feel for the accuracy and variations that might be achieved with Orchids. I’m looking forward to spending more time experimenting with various descriptors. I haven’t yet looked into filtering the spectral features. I’m wondering if our Orchids Gurus can make some suggestions, in the way of heuristics, on how these might be best used. My intuition tells me that one useful case would be in using a filter to isolate the attack and sustain portions of a target analysis, i.e. filtering out the beginning of a temporal spectral centroid descriptor.
Salut,
Kenneth.
kenneths-kenong-experiment.zip (4.19 MB)