Why Piano Response Matters When Practising

There is no question that preparing an interpretation for a concert is easier when the instrument you practise on sounds as close as possible to the concert grand you will eventually perform on. In this respect, sample-based virtual instruments have a major advantage: the enormous number of individually recorded notes allows them to reproduce the sound of an acoustic grand piano in great detail.

But for practising, it is not enough for an instrument simply to sound similar. When working on a demanding piece, we are not just learning which keys to press and when. At the same time, we are learning very subtle relationships between our movements and the acoustic result: How much force do I need to produce a certain dynamic level? How does the timbre change with it? How far can I bring down an accompanying voice? And how immediately does the sound follow my movement?

If the practice instrument behaves very differently from the concert grand in these respects, we have to readjust, at least to some extent, when moving to the concert instrument. With Live Grand, this is exactly what we want to minimise. Our goal is therefore not only to reproduce the sound of a Steinway D. We also want the relationship between finger movement and the resulting sound to resemble the acoustic original as closely as possible. Two properties are particularly important for achieving this.

Latency

The first is latency. In his Key Features video, Paul mentions a measured total latency of less than 10 ms – from pressing the key to hearing the resulting sound. Achieving this required a considerable amount of optimisation. This is why we consider low latency to be one of Live Grand's key features. Without appropriate optimisation, the key-to-ear latency of a virtual instrument can quickly exceed 10 ms, even on good hardware.

Why differences on this time scale can matter to a pianist can be illustrated very nicely with Chopin's Op. 66. Yundi Li won the International Chopin Competition in 2000. In one of his recordings, he plays this piece at approximately 190 bpm.

Let's look at the 4-against-3 polyrhythm in bar 5.

Mathematically, this pattern can be divided into twelve equally spaced positions. This gives us intervals of approximately 26.3 and 52.6 ms between successive notes played by the two hands. A musical interpretation, however, does not follow this mathematical grid exactly. An analysis of Li's recording shows that individual positions can shift by approximately 10 to 20 ms relative to this grid.

So when we talk about interpretative microtiming, we are already dealing with the same order of magnitude as the latency of a virtual instrument. Of course, this does not mean that 10 ms of latency automatically results in 10 ms of timing inaccuracy. Pianists can adapt to a constant delay to some extent.

For our goal, however, something else is important: If practising is supposed to feel as similar as possible to playing the concert instrument later on, we also want to alter the timing relationship between finger movement and acoustic feedback as little as possible. Even if a pianist adapts completely to an additional delay and ultimately plays just as accurately in objective terms, the instrument may still feel less immediate or less “connected” to the fingers. And when switching to the concert grand, that additional delay disappears again.

This is why we have put considerable effort into keeping the additional latency introduced by digital sound generation as low as possible.

Dynamics, Timbre and Touch Calibration

But the timing of the acoustic feedback is only one half of the equation. Just as important is what acoustic result a particular finger movement produces. Let's return to the same 4-against-3 polyrhythm in Yundi Li's performance.

The left hand is remarkably quiet here. This is also exactly what pianist Kazurou Ooi recommends in his analysis for the PTNA Piano Music Encyclopedia: apart from the bass notes, the left hand should be played almost ppp. As a result, it barely pushes itself into the foreground. Over this calm foundation, the right hand is free to accelerate, pull back and “breathe”.

The complex polyrhythm is therefore hardly perceived as a mechanical “four against three”. Instead, it creates this distinctive sense of floating. If both hands were equally prominent rhythmically and dynamically, we would be more likely to hear two interlocking mechanisms rather than one coherent musical movement.

Passages like this demonstrate why simply recording a concert grand at as many different playing intensities as possible is not enough to create a realistic practice instrument. On an acoustic grand, a stronger attack does not merely increase the volume. It also changes the timbre. Volume and timbre are therefore two interconnected results of the same finger movement.

The different sample layers in Live Grand reproduce these different states of the original instrument. But for them to behave correctly while playing, we also need to know which finger movement on the keyboard being used corresponds to which state of the original grand piano. This is where touch calibration becomes important.

Live Grand not only reproduces a dynamic range of up to 55 dB. We have also measured how different playing intensities on the original grand affect both volume and timbre. We then measure selected keyboards and calibrate Live Grand to their individual touch characteristics.

Different MIDI keyboards do not necessarily produce the same MIDI value from the same finger movement. With an arbitrarily chosen velocity curve, certain dynamic regions can therefore become too wide or too narrow, or occur at a different point within the touch response.

Let's return once more to the extremely quiet left hand in our Chopin example. For us, it is not enough that Live Grand is simply capable of producing this very quiet sound. Ideally, the pianist should also be able to produce it with a degree of touch that is as close as possible to what would be required on the original grand piano. Otherwise, while practising, the pianist might learn to play the left hand very quietly – but potentially using a different movement from the one required later on the concert grand.

And this is also where sample layering plays an important role. Different keyboards resolve different areas of the touch range with different levels of precision. We therefore do not simply distribute the recorded sound layers across MIDI velocities from 1 to 127 according to a fixed scheme. Instead, Live Grand takes the measurements of the individual keyboard into account when assigning both volume and sample layers.

This gives a particular touch range exactly the amount of space on the MIDI scale that the respective keyboard needs in order to reproduce that range as precisely as possible. The objective is always the same: When the pianist changes the finger movement, volume and timbre should change together as closely as possible to the way they would on the original grand piano.

CONCLUSION

This also explains why we consider these particular characteristics to be key features of Live Grand. When practising a demanding piece, we are constantly training the relationship between a very precise physical movement and the sound that results from it.

Latency affects when we hear the result of that movement. Touch calibration determines how our finger movement is mapped onto the dynamic range of the original grand piano. And the corresponding volume and layer mapping determines how volume and timbre develop together.

If these relationships behave very differently on the practice instrument and the concert grand, we have to adapt accordingly when switching instruments. Live Grand therefore does not simply attempt to reproduce the sound of a Steinway D as accurately as possible. Our goal is for the relationship between finger movement, timing, dynamics and timbre to resemble the acoustic original as closely as possible as well – so that what you develop while practising carries over as directly as possible when you move to the concert grand.

Authors

Dominik Schulte is a pianist, piano teacher, software developer and co-founder of Live Grand. Combining his practical experience as a pianist with his background in computer science and AI, he developed the technical concepts behind Live Grand, including precise keyboard calibration, low-latency performance, advanced piano modeling and AI-assisted sample processing.

Paul Bittmann is a freelance composer for film and games with a background in computer science, audio systems, music implementation and sound design. He studied composition at Berklee College of Music and works with Dominik on the development of Live Grand.

How Low Latency and Touch Calibration Improve Piano Practice