One of the original methods for producing digital piano sound was that of ‘sampling. The principle behind sampling is a simple one: if you wish to have the sound of a middle C play when you press that key down on a keyboard, then record a real piano playing that note. Then, when you press it, it’s like pressing ‘play’ on that recording – you hear that note just as it was played on the real piano. Single samples refers to a configuration where each note has had an individual recording made of it, at one dynamic level. For entry-level units, sometimes a single sample is assigned to clumps of notes, and that sample is then pitch shifted up or down a few tones to match the note. This was considered a major innovation a few decades ago when it first entered the market, since it allowed for a real recording to be played back for each key. However, to accommodate the many dynamic (volume) ranges that pianists demand, there were complex algorithms developed to modify these single samples to better. After all, when you play a real piano very aggressively, not only the volume but the tone itself is changed – the more authentic a digital piano, the more this change is properly represented. This method has now been superseded by both multi-layer sampling and modelling.

Multi-layer sampling operates on the same principles of single-layer sampling, however several recordings are made of EACH note on the piano, each one made at a different volume level. In a studio setting, software pianos are able to render as many as 16 sample layers, each one recorded at a different volume, for EACH key (note: this is not to be confused with volume levels per key, which is a measure of how accurately a digital piano can respond with a volume adjustment proportional to the players mechanical input). On stand-alone digital pianos, the limit is much lower, with the very best in the industry offering a few. The exclusive reason for the multiple layers? Give more authenticity to all dynamic (volume) levels that a pianist might require, so that you don’t have to try and change a sound with programming if they play very softly or loudly.

Modelling is completely different than sampling, and produces quite a different digital piano sound. It does not play back a recording of a piano. Using very fast computing not available 10 years ago, it actually synthesizes the sound based on dozens of parameters. Think of it as a real-time computer model of a piano. In some ways the technology is not new in any way: university labs and very high-end music studios have been synthesizing incredibly complex sound for decades. The problem is that the more complex the sound, the longer the processing time is. If you took the piano model used in Roland’s V-Piano for example, or Kawai’s MP10, and fed it through the common computer processor of the day, but from 20 years ago, it would take nearly a minute for the sound to come out. In otherwords, you couldn’t play it in real-time.

Because processors and computer hardware is so inexpensive today compared to it’s power, most companies in the electronic instrument industry have focused their efforts on the modelled / synthesized method vs. the sampled, although it will still be years before this high-end technology filters down into the main-stream of the market; at this point $3500 seems to be the entry point into modelling in a stand-alone unit.