Roberto Gordo Saez wrote: > I've noticed that the disklavier is recorded very dry (probably on > purpose), so adding a reverb is almost essential; otherwise it sounds > like if the head is put inside the piano :-) William Coakley piano > samples contain some ambiance, and I guess they are recorded in real > stereo (unfortunatelly this can't be done in the disklavier soundfont, > since source samples are in mono). Yes, I believe that the Coakley samples were recorded in stereo. One of the things that impresses me is that although William Coakley is an engineer, he is also a musician (he plays piano himself). This made him approach the project in a different way, focusing in it as much through a musician's eyes (and ears) as through those of a sound engineer. This brief article of his about his observations and approach is very interesting: Why Bigger Isn't Better -- by William Coakley http://williamcoakley.com/articles.php?article=bigger.php An analogy I might make myself is that it is not necessary to photograph a scene simultaneously from 5 different angles or perspectives in order to get a good photo of that scene. I personally believe that one could create a smoother, more expressively responsive piano soundfont with fewer layers, with perhaps only 2, or 3 layers at most. Coakley's piano samples, to the best of my memory, have only 2 layers, one strongest at minimum velocities and the other strongest at maximum velocities, with several variations in the gradation between them (several sound "patches" or settings of the same samples). One reason I believe that fewer velocity-layer samples would be needed is that in a piano, perhaps in contrast with a few other musical instruments, the basic character of the sound--to the ears of a person or musician, not necessarily to an engineer with a waveform analyzer--doesn't change character that much. Yes, there is a difference in sound when a piano is played softly and when the keys are depressed with great force, but the intermediate volume levels and characteristic sound are just a relatively smoothly graded variation from one to the other. In the Disklavier soundfont, it seems (I'm not sure) like the forte samples predominate. In order to accommodate all 5 layers, it seems that samples for ffff, fff, ff and f are spread out too much over the velocity spectrum, infringing on the space that should normally be reserved for softer layers. This would seem like a natural thing to do, because the alternative would be to allot a very small area of velocity values to each of ffff, fff and ff (or fff, ff and f--each of the strongest-sounding samples), which almost makes one or more of those samples seem superfluous (which perhaps they are). :-) If a person considered the dynamic range of a piano as going from soft to loud as follows--ppp, pp, p, mp, mf, f, ff, fff, ffff--then I would think that samples of "p" and "ff" or "fff" would be sufficient to represent the entire dynamic range, in just 2 layers. This is of course just my opinion. Also in the Disklavier samples, were the pp and p (soft) samples normalized to raise their volume? I ask this because the volume or dynamic range of the soundfont seems compressed. It almost sounds as though--despite triggering the sounds with various velocities--as though the soft samples have been amplified or normalized and the loud samples have been compressed. In other words, it seems difficult to achieve the full dynamic range of a piano when playing the soundfont with a full velocity range on a MIDI keyboard (even when setting the MIDI keyboard to different velocity-response curves). These are just questions and notes intended as "input and feedback" only. (Hmm-- "notes," "input," "feedback," all musicians' terms :-) Best wishes, Steve _______________________________________________ Linux-audio-user mailing list Linux-audio-user@xxxxxxxxxxxxxxxxxxxx http://lists.linuxaudio.org/mailman/listinfo/linux-audio-user
