> There's something seriously lacking in psychoacoustics in terms of > estimating the structure of psychoacoustic spaces. In the case you > mention, Andy, the timbre spaces are (relatively) high-dimensional,
Hmm, those *are* interesting subjects, but in terms of the "normal" connection of instruments with sounds, in (physically sounding) spaces, with human perception limits and emotional connections with chords and envelopes, in music still designed to be listened to instead of merely blasting people away, there are many theoretical considerations which can keep us out of the woods of mystery and probably a good thing or two which can be said about the value of particular types of interpolation.
The main idea of an instrument or even a sample being played in a reverberant space, or even a sound effect sounding through lets say an ipod is not correctly summed up easily in either strict time domain or strict frequency domain. Even when using accurate samples (accurate recording in the time domain of the pressure waves) to preserve an instruments' or sounds' "sameness" over various speakers and amplifiers and such will require accurateness normally not available. Even so called studio monitors and decent enough HiFi amplifiers will unfortunately reveal more often how they combine certain types of distortion with certain averaging effects than be suitable for a neutral "web-test".
Frequency-wise probably one of the main considerations for "equality" of sounds is the perceptual loudness: the well known reverse-bell type of equal-loudness perception dictate that sounds which are alike may well sound different when played at different volume. Many sounds which interest sound designers and producers alike have a clear aspect of loudness-sensitivity built in them.
And that is without considering the technical details of the limitations of certain ways of doing things: what is the effect of the length of an FFT interval, when (correct) wavelets are being used how many and how long (and orthogonal) are they. When filters have been used: what (electronics) type are they, or which were the motivations for trading off the various impulse responses being used? Was the sampling Nyquist issue honored properly, are the rules of envelopes and transients in the frequency domain computations properly honored? Was the reconstruction filter (and it's habitual stringent limitations) considered in the listening test ? Are the considerations about sound like-ness convolved with the listening space, either the recorded (or synthesized) reverberation or the properties (esp. resonances) of the listening space, and what is the intention of such effects (small market for exact reproduction or broad acceptability of the resulting sounds) ? I mean in the history of modern music these are HUGE subjects.
Of course general (universal) science can stoop to the level of wanting a techno-type of approach about such huge subject, considering a small set of options or going for a particular small subject, and build a whole theory around that. Honestly, the "equal-sound" spaces and their orthogonality will probably best be found in pro audio circles with quite heavy equipment, and good convolvers of those types of dimensions are probably called names like Lexicon (famous since the beginning of the 70s for deep Digital Signal Processing). Picking out "sound-interpolation" in that world is either player-ish (like a piano on a pad-computer) or a sign of higher science being attempted.
*Making* interpolated sounds of quality of course contains a major technical component, which when researched in depth leads to profound requirements for the research candidate concerning signal processing. of course it is tempting to hide where the "universal" science hides and what the personal technical skills and used (abused) work of others lurks in the subject of for instance a thesis. In the work at hand I found the various questions answered too unsatisfactory at the level of a small local "uni"versity working on interesting subjects (too popularistic), the level of the musician (virtually non-existent), at the level of DSP too much was not quality enough or theoretically to unsolid in it's foundation (as appears fashionable), and at the level of the general interest my EE senses weren't satisfied, nor my "advanced science" taste buds (I could interpolate FFT based synth sounds on an Atari ST already ages ago.
The two main time and frequency considerations I mentioned combined, have people deluded all the time. By lack of widely available teachings on the subjects, it may well be better to aim for university level teachers who fit the profile of a solid researcher, rather than people who probably would answer a vocation tests' psychological profile section by associating too much with dungeons and murder weapons combined with desires for activities with women tired of life...
Theo Verelst -- dupswapdrop -- the music-dsp mailing list and website: subscription info, FAQ, source code archive, list archive, book reviews, dsp links http://music.columbia.edu/cmc/music-dsp http://music.columbia.edu/mailman/listinfo/music-dsp
