< Back to IRCAM Forum

Audio setup and descriptors — a few questions...

Orchids
#1

Cheers everyone,

First of all, congratulations to the team for making this beautiful piece of software.

I’m a near total neophyte to sound analysis and synthesis. My questions are as simple and naive as I’m guessing the answers will be subtle and nuanced:

What would be the descriptors best suited to evaluate solutions in terms of sensory/musical consonance? I’m as interested in orchestrating sampled sound generally as I’m interested in reorchestrating my music with Orchids.

When reorchestrating my music what should I be looking for as audio and descriptor setups (in function of the target’s instrumentation, tempo, etc.) in order to preserve as much as I can the melodic, harmonic, and rhythmic structures of the target (input)?

I’ve read the available Orchids’ documentation, and already tried auto-pilot and multi-targeting with both monophonic and polyphonic music targets and the results I’m getting are hardly congruent with the target (with Harmonic Filtering on or off).

Any clues and guiding lines on these issues would be much appreciated.

António

#2

Dear Antonio,

Thanks for the support. In fact, there is no “direct” solution to your question (as you inferred correctly). Usually the descriptors highly depend on the target you input and most of all on the aspect of sound that you want to reproduce. We usually advise to rely on the “PartialsAmplitude” (or PartialsMeanAmplitude for static targets) and “MelAmplitude” (MelMeanAmplitude for static) for best results. However, these are quite holistic descriptors. Depending on targets, each and every descriptor will give you a rendition of a different “aspect” of sound. We will update the documentation this week to give a complete description of every features provided by the system, so that you can choose with better intuitions on the corresponding aspect.

Regarding the “congruency” to the targets, there is two things to know :

  • The “multi-target” is a very alpha and primitive mode, and we strongly advise against using this mode if you want precision on the optimization, as the selection of best solution is just a simple mean, without any subtlety … Hence, you should attain better results by optimizing several chunks of your target (using the selection window in either the target or the partials windows), and then selecting your preferred solution for each chunk and then constructing a “maquette” by hand.
  • We spotted through the user returns, that there is still a major flaw in the Partials computation, which explain the lack of correlation in this optimization. We are working very hard to fix all the corresponding issues, and will be making a new complete release (0.9.1), that will fix all of this before the end of the week ! (and should provide a widely higher quality of optimization)

All the best,
Philippe

#3

Hello António,

There is a paper with comprehensive technical description of the descriptors and some discussion of usefulness here:

http://recherche.ircam.fr/anasyn/peeters/ARTICLES/Peeters_2003_cuidadoaudiofeatures.pdf

What would be most useful is a kind of taxonomy of sound objects with mappings of the salient descriptors that best describe that kind of sound. i.e. My intuition is that, for a sound with a mix of noise and harmonic elements, one useful descriptor would be signal zero crossings as that describes a difference between noisy and periodic aspects of the signal (frequent zero crossing indicate noise, infrequent crossings periodicity). I’m interested in finding out from others what mappings are most useful for what kinds of sound targets.

I think spending time in the Explorer page of Orchids looking at the graphs of the various descriptors for different instruments is a good way to start sorting these out. Select an instrument sample from the Explorer panel to the left, choose a descriptor in the Inspector panel and watch the relationship between the cursor and the graph of the descriptor as the sound plays. As an example, you can confirm the zero crossing descriptor’s information I described by looking at the Harp bisbigliando-with-stick samples: Zero crossings start out very high while the stick is hitting the string, then drop down as the string is allowed to vibrate off in it’s “normal” periodic way.

The potential of this forum will be to provide this kind of practice-based knowledge in finding creative uses for Orchids. I’m looking forward to learning from everyone!

Best,

Kenneth.

#4

Thank you both Philippe and Kenneth for the tips and eye openers, it was very helpful.

António