All about “software pianos”

Hi everyone! It has been a busy summer, but I am finally back to write about virtual-piano software programs or, as I will refer to them, “software pianos”. So, shortly after I got my digital piano and was enjoying all the benefits I’ve been describing on this blog, I started lurking on the forums at PianoWorld.com — specifically their digital forum. I noticed that many people there were using software pianos in combination with their digital pianos. I decided to do some research, and I am sure glad I did! I’ve delayed writing this post because I’ve really been going to town playing around with this stuff, and I wanted to be really well versed on the subject before sharing my thoughts. This post will be an introduction to software pianos. I will explain what they are, I will mention certain ones that are well-liked, and I will share some general thoughts I have. Future posts will describe what I really think of each one I’ve tried and will include sound clips I’ve created with them. At some point I will wax philosophical on what software pianos can do for a professional classical pianist such as myself.

What is a “software piano”?

A software piano is a program that you run on a PC (either as a standalone application or within a “digital audio workstation”). You connect your physical digital piano to the computer running the software. (Digital pianos have either MIDI or USB ports that with the proper cable can be used to connect them to a PC.) Then when you play,, the software piano will generate sound. You turn down the volume on your actual digital piano, and listen instead to the audio produced by the computer. You can connect headphones or a speaker system to the computer.

Why use a “software piano”?

The reason is that most software pianos sound significantly better (more realistic) than the audio produced by the digital piano itself “on-board”.

How do “software pianos” work?

Now we get into the interesting part! There are two kinds of software pianos: those that are “sample-based” and those that are “modeled”. The on-board audio produced by most digital pianos is “sample-based” as well. Here goes an explanation as best as I can manage of what these two terms mean and how they are applied to the process of making music digitally.

How “sample-based software pianos” work:

Most decent digital pianos and all sample-based software pianos use actual recordings of an acoustic piano to generate their sound. The basic process is to:

  1. Set up a nice grand piano in a good recording studio with microphones in several different locations
  2. Record each note of the piano (or at least several notes) at several different dynamic levels
  3. Record the notes with and without the pedal down
  4. Record what various note releases and pedal releases sound like
  5. Use programming to serve up these various recordings on-demand super quickly
  6. When you fire up your digital piano or software piano and interact with the keyboard, the recording that corresponds to what you are doing is played

Now, obviously, it’s not as straightforward as that because so many variables go into a piano’s sound that not every single possible sound made by a piano can be recorded. To even come close, would take so much computer memory that it wouldn’t be feasible at all. So, all “sample libraries” (the collection of recordings used by a digital piano or software piano) have to take shortcuts to a greater or lesser degree.

Here’s an example of the kind of shortcut I’m talking about. When you play and hold down a piano key, there is an initial sound called the “attack”, the sound made while the key is held called the “decay”, and the sound made when you let go of the key called the “release”. Attack sounds quite different at different loudnesses. There is more than just a difference in volume. There is a difference in timbre; you get more overtones the louder you play a key. So, one of the first decisions made by someone “sampling” a piano for digital use is how many volume levels to record the attack at. This is referred to as the number of “velocity layers”. Digital pianos will express the force you play the key with (the velocity with which the key goes down) using a number from 1 to 127. That number is sent to the computer or the digital piano’s own built-in processor, and a sound sample is triggered. But, there is no software that uses 127 discrete attack samples per key. That would be absurd and unfeasable. This is one area where the differences between a digital piano’s on-board sound and a software piano become apparent. The faster the processor, and the faster the disk reading speed, and the more disk space you have, the more samples can be stored and served up on demand. Since a digital piano’s brain isn’t nearly as powerful or sophisticated as a high-end PC’s, the digital piano’s built-in sound library might use 4 velocity layers per recorded note and then just interpolate between them for the other 123 possible volume levels. In contrast, the best sample-based software pianos have around 20 velocity layers. The result is more realistic and smoother changes in timbre as you play the keys softer or louder.

Here’s another example. A digital piano likely won’t have discrete samples for all 88 keys. Rather, it might use recorded samples for every 4th key and then use programming to interpolate the keys in between. Software pianos usually sample all 88 keys individually. Again, the result is more realistic.

One more example. The decay of a piano note is a very complex thing. It’s not like the sound made by the attack just continues and gets quieter over time. No, no, no… It’s much more complex. Because there are 3 strings for most of the notes, and a soundboard, and countless other things that impact the sound, as a note decays, you get a whole series of overtones that come in and out of phase with one another. Like I said, complex. If you really listen carefully to what a piano note sounds like as it decays, it is just fascinating. A digital piano will use a sample for a part of the decay for each velocity layer, but since the decay of a piano note can last a very long time (up to 2 minutes for a bass note of a concert grand), at some point, the sample is cut off and then “looped” or played over and over again while the volume is decreased. Doing this uses less disk space and less memory than playing a long sample of the whole decay so it is a necessity for the digital piano’s on-board sound. Unfortunately, this results in a very unrealistic decay. It’s one of the few things that really *bothered* me when I started practicing on my digital. I could live with it, but the longer I held a note or chord, the faker it would sound. It was very noticeable to me. However, some of the best software pianos actually *do* sample each note’s complete decay for each velocity layer. You have to have a pretty powerful computer to run that kind of software, but the difference is really amazing.

Holy moly, this post is getting long………

Anyway, that basically explains what a sample-based software piano does. In summary, though sample-based software pianos use essentially the same technology as the built-in sound generator on a digital piano, because they run on a powerful PC, they can produce sound with greater realism and fidelity. There are limits to what they can do, of course. I didn’t even get into how differently notes sound with pedal down vs. pedal up or how differently they sound when played with other notes vs. alone. You get the picture. Like I said, it’s not possible to accurately record every possible sound a piano is capable of making and then serve up those sounds on demand. But, most sample-based software pianos are *much* better than a digital piano’s on-board sound, and if you make a recording with one, it is hard to differentiate between it and a recording of an acoustic. After listening for a while, most musicians would still be able to tell it’s not a traditional recording. (Most “normal” people probably wouldn’t be able to tell.) Each sample-based software piano sounds very different due to the *many* factors involved (and they all have many options in their interfaces to alter their sound in various ways), and it is great fun to compare and contrast them. If you are interested, I recommend you look up some of the most popular ones on the internet and check out their sound clips and descriptions. Good ones to look at include: Synthogy Ivory II, VI Labs True Keys, VI Labs Ravenscroft, Garritan Abbey Road CFX Concert Grand, East West Quantum Leap Pianos, Vienna Symphonic Library Vienna Imperial.

How “modeled” software pianos work:

Oh boy, now as hard as it was to describe sample-based technology (and who knows if I did a good job), this is harder. I’m not sure I really understand it myself, but here is my best attempt at an explanation. In contrast to the sample-based software pianos I described above, modeled software pianos don’t use *any* pre-recorded samples. They use sophisticated physics and mathematics to create a piano sound, from scratch, in real-time. It’s *all* done with computer programming. So rather than being a computer program written to serve up and interpolate between actual audio files, this is a computer program that understands all the factors that go into making a piano sound the way it does and is able to create the sound on the fly.

I’m not sure I can do better than that explanation. To me, it is a miracle of software engineering. There are a few digital pianos that use this technology to create their on-board sound, and there are a few software pianos that use it. The best modeled software piano by any measure is the newest version of PianoTeq. This is the software I have been playing with rather obsessively for the last two months.

The benefit of a modeled software piano is that it needs only a fraction of the disk space and computing power needed by the sample-based ones. Because computers are designed to do math really fast, the modeled software piano is a perfect fit. It runs better, with less latency (delay), and can account for a lot more of the factors that go into a piano’s sound than a sample-based software piano can. The programmers can infinitely tweak the algorithms that produce the sound to allow for all kinds of variations. This makes it possible to produce a wide variety of piano sounds for different purposes. PianoTeq allows the user to fiddle with a nearly infinite number of variables to try to get a sound you love. It is more versatile, more light-weight, more responsive, and more nuanced than any sample-based software piano all because it does *not* rely on a set of of fixed, pre-recorded samples. I’m in awe of what it can do. Please read the first comment below this article by Charles Cohen for more detailed explanation of this technology. And, as he says, check out the info at pianoteq.com.

If you are still with me, I thank you! I am going to wrap this up now. My next post will detail my own experiences with the software pianos I have tried, and I will post sound clips!

in Digital Piano Tech

11 Responses to All about “software pianos”

  1. Charles Cohen says:

    That’s a very good explanation of sample-based software pianos.

    I’d explain modelled pianos differently, FWIW. This probably goes into more detail than you wanted — it’s about “one level deeper” . . .

    A “model” is a mathematical representation of a physical object. A single piano string has a fairly simple mathematical model. The model can predict (reasonably well) what happens when that string is hit by a felt hammer. It spits out (as its result), the vibrations of the string as it builds and decays. And — because it’s a model, not a recording — it’s fairly easy to modify things like:

    . . . How hard is the hammer?
    . . . Where is the hammer located, along the string?
    . . . How long, and how thick, is the string? Is it wound, or bare?
    . . . How much energy goes from the string, into the piano’s soundboard?
    . . . How is a string affected by the vibrations of other strings?

    Each note on the keyboard has one, two, or three strings to model. The vibrations of the strings couple into the air (a little bit) and the soundboard (a lot). You need a model for the soundboard, a “tweak” to the string model to allow for the dampers, and so on — all those add complexity to the overall “piano model”.

    Then, there’s the way that the soundboard vibrations couple to the air, and feed into the microphones, which gives another level of freedom:

    . . . Where are the microphones, around the piano?

    The “miracle” is that a desktop or laptop PC has enough computing power to figure all this out, _as you play, with no significant lag_.

    So the same software can “sound like” many different physical instruments — big grands with soft hammers, small grands with hard hammers, uprights of various heights, and so on. Because of the approximations of the model (it’s good, not perfect), it won’t sound _exactly_ like a real instrument. But it can get reasonably close.

    Anyone who needs more depth can explore the Pianoteq website, which has good background material.

    Your work is doing a lot to validate what us amateurs have been hearing. Thank you!

    • Thanks so much, Charles. I have added a suggestion to read your comment to the article. In future points I am going to go on and on about PianoTeq, so I sort of cut my description short. I really like how you explain it, though!

  2. pwl says:

    The clarity, directness and ease with which you write about the oftentimes amorphous subject of software pianos is refreshing – and appreciated. This post should be a required “first read” for anyone wanting to understand or use software pianos – or make comments in online forums about them!

    • Thanks! I looked and looked for something like a good “first read” on the subject and couldn’t find anything detailed enough but not too detailed. So, I wrote it! Glad you like it.

  3. Jan van Dijkhuizen says:

    Hi Rachel,

    Great posts; thank you! Glad to hear you’re enjoying Pianoteq.

    Another area in which Pianoteq is so much better than the sampled competition is in the interaction between strings — i.e. sympathetic resonance and its big brother, sustain resonance. I’ve yet to hear a sampled piano that does these things really well, and, as you probably know better than I do, it’s an essential component of the sound of a piano.

    Pianoteq is very good at this and it’s one important reason why the experience of playing Pianoteq is so much more satisfying than playing a sampled piano. Physical interaction between strings is inherently difficult to capture in samples because it’s not static.

    Try playing the final 45 seconds or so of Chopin’s prelude in F# minor (or play all of it! I can really only manage the last bit). In Pianoteq you can really ‘soak’ in the rich wash of sound created by all the overtones. On TrueKeys, Synthogy Ivory II that just doesn’t happen; the notes don’t ‘gel’ and you don’t get that flowing Chopin sound. In my experience it’s less of a problem with, say, Bach.

    Anyway, hope this is helpful. Good luck with Pianoteq!

    • Thanks for that perspective!

  4. Claudio says:

    Perhaps beyond the scope of your research but if you like the results from modeled piano software it would be valuable to compare those results to a modeled digital piano – Roland calls their modeled piano internals ‘SuperNatural’ and their DPs that use it start at $899.

    • Thanks, Claudio! If Roland would like to send me a unit to test, I’d be happy to check it out and blog about it. 🙂

    • DigitalGuy says:

      Supernatural is not a modelled sound engine, but a sampled-based piano with 3 to 4 velocity layers and modelling filling in the gaps for 128 total levels of expression. Also, all the resonances are modelled and therefore controlled dynamically. This gives a sound that, IMHO, competes with some of the best software pianos. Examples of purely modelled DPs include the V-Piano and the Physis piano. Whether they are better or not than the current Pianoteq is a matter of personal opinion (in my opinion they still are in several respects). They are certainly much more expensive, however…

    • Thanks! I had a feeling I’d learn a lot from my commenters. 🙂

  5. Pingback: Chopin on a digital??? | Fundamental Keys / Rachel Iris Jimenez

Submit a comment

Your email address will not be published. Required fields are marked *