I've analyzed several wave files and recorded some PianoTeq notes to wave files to compare the results. What I found was that the wave files from all of the sources varied widely, and that the PianoTeq files did not have qualities that were peculiar to PianoTeq.
All I can say for certain is that the major differences in the various wave files lay in:
1. The number of what I'll call "near frequencies" for each note. In other words, some wave files of individual notes tended to include more frequencies that were neither the fundamental nor the partials, but were instead frequencies just above and below both the fundamental and the partials. (So an A file at 440, for example, would include the frequency at 440, of course, but would also include, at fairly loud levels, freqs of 438 and 443 or so. The lower partials would also have these near frequencies. I've sometimes thought that these frequencies are those of the slightly detuned unison strings. However, they may not be--some analysis programs (such as GoldWave) show a slope of frequencies that peak at the fundamental & at the partials, instead of the distinct lines for frequencies. They may instead occur because a piano string "emits" a broader spectrum of frequencies than an abstract Fourier formulation accounts for. PianoTeq includes this broad range of frequencies instead of the distinct near frequencies, and so do many sample libraries. I was surprised, actually, that PianoTeq didn't have more of them, even when unisons are detuned. But many sample libraries also lack these. (My ZR76, with the Perfect Piano samples, for example, has few of these distinct near frequencies, but has an obviously different timbre from PianoTeq.) Regardless, I did find that the few notes that had these near freqs (instead of just a wide band of freqs near the fundamental and the other partials) sounded full and rounded--a few samples from a Yamaha P-200, for example, had them, along with the wide band of near frequencies. But I don't think I miss them in PianoTeq---each note sounds good to me. Sometimes I think that it's possible that when chords are played, their absence is more noticeable--their combined effect is missing? Particularly in the octave above the middle C octave? At other times, PianoTeq is so convincing that the addition of other frequencies would be bad. On the other hand, what I'm seeing may be more the way different analysis programs display the analysis: some may just take the loudest near frequencies and display them as separate bands. Or those that display a wide band of frequencies near each fundamental and partial may be "imagining" the freqs between--really only a few freqs are there, but the display shows a slope or a colored area instead of distinct bands. (I'll ask about this difference in displays in a separate topic, come to think about it.)
2. Some wave files of notes at the same velocity have more "warbling" frequencies than others: by these, I mean the transient sounds, occuring more at high velocities but also at low velocities, and changing pitch over the first few milleseconds. They create a wiggly up and down line on the analysis programs. (When listened to in isolation, frequency by frequency instead of as a group of transients, they warble.) Pianoteq has these and seems to increase them, as it should, as higher velocities are reached.
3. The initial frequency of the fundamental is slightly sharp in many sample libraries and in Pianoteq. A few libraries, for reasons I don't understand, don't seem to have as pronounced a raised frequency at the start. Possibly they were assigned midi velocities that were higher than their actual strike velocities.
4. The amplitude of some partials in some wave files varies slightly, dipping to below audible level and then rising back up so they can be heard, in the first 2 seconds or so. In other sample libraries, the amplitude of the partials only declines. I've seen one theory that explains this phenomenon as the sound of the notes bouncing off the interior of the cabinet and coming back to the mic. (I don't know my source for this. Sorry.) Other people argue that the fading in and out of partials is instead the "natural " evolution of a piano note. I don't see a lot of this variation in PianoTeq, but I'm not sure if that's good or bad.
5. The frequencies of some lower partials in some samples varies slightly in the first few seconds after the note has stabilized--after the attack portion, in other words. Not enough for my ear to register a change in pitch, but the freq does seem to shift. I couldn't tell any change in the "quality" of the note. Most samples, and SampleTeq, do not share this change in pitch.
These are obviously rough observations that come more from loading up some files at night and briefly comparing them than from a systematic study. I'd be interested, however, in learning if other people come up with similar findings using whatever piano samples they have. The thing that interests me most is the number of near frequencies, although I'm not entirely sure that they would make a good contribution.
Last edited by Jake Johnson (09-01-2008 18:52)