by Eric Keller
One of the most remarkable developments of recent years has been the arrival of virtual instruments. In this contribution, I want to compare the simulation of a number of pre-piano instruments available in the Pianoteq, Garritan and EastWest offerings.
I rush to say that I have absolutely no ties to any company mentioned here. I'm simply writing as a person familiar with some of the complexities of instrument simulation and as an absolutely delighted user of these virtual instruments in my daily work.
Before diving into the comparisons, it is probably a good idea to explain three main approaches to the virtual simulation of musical instruments.
The Concatenation Approach (virtual orchestras)
The most computer-intensive approach is to obtain recordings of all sounds played by a given instrument in various combinations, such as a G played in the context of a rising or falling second, third, fourth, etc. This is the approach used in the well-known Garritan or EastWest systems. This sound acquisition process has to be repeated at different playing speeds. From all these recordings, the sound engineer extracts sound chunks that can be chained together ("concatenated") to recreate new music in the appropriate contexts. Since these sound chunks are recorded at very high-quality sampling frequencies, this approach creates huge data bases. The EastWest data bases begin at around 11 Gb and can easily go up to 200 or more Gb for a full orchestral set.
One of the net consequences of this approach is that on consumer-level computers, real-time performance is often impossible when using these large data sets. There is a difference between Garritan and EastWest. On Garritan, simulation proceeds offline. It can easily take a quarter hour to obtain just a few minutes of simulated musical performance with a complex orchestra. On EastWest, there is some extra smart engineering to read in the data at the necessary speed -- however much time that may take -- and then to reproduce it in real time, as long as you have a fast computer. On slow computers, sound reproduction "hickups" may still occur.
These time limitations can be frustrating, e.g. when performing on-stage. But currently, only this type of concatenative synthesis provides synthesis for all commonly used instruments in high-quality simulation, and it is the only way to go when multiple instruments must be combined into an ensemble or an orchestra on a single computer.
The Sound Font Approach (notation programs, electronic keyboards)
Obviously, real-time simulation is crucial in musical composition and notation programs, and it is simply required for electronic keyboards. Out of this need has arisen the soundfont approach. Here representative sound chunks are stored in much smaller data sets and are reproduced exactly when needed. The better soundfont data bases include different versions for long and short versions of the sound as well as various forte and piano versions.
This simplified approach has been used in notation programs for many years now, and it has been incorporated into electronic keyboards. With the increase of the speed and capacity of computer components, better data sets and more extensive soundfont bases have become possible. My favourite soundfonts for my notation program (MuseScore 2) are the full FluidR3 GM2, Musyng Kite and RealFont sets.
The Parametric Approach (high-quality daily practice with small footprint)
But there is more! For a long time I had been disappointed by the shallow sound of my electronic keyboard and by the limitations of soundfonts. Surely, one can always trade up to the latest high-class keyboard and try out new soundfonts. But there is also another solution. I found out that the French sound engineers at Modartt have taken a much more sophisticated approach to virtual music simulation.
Concentrating on keyboard instruments, they have created simulations of a few high-quality instruments by determining frequency profiles for each of the sound combinations mentioned above. Then they observed how these profiles changed as a result of a large number of parameters. They took into account all relevant instrument parameters, such as the action of the hammers and dampers, the profile of the sound board and of the strings, the resonances within the instrument, the noises from the pedals and of the action mechanism. Other parameters were room reverberation and possible echos, as well as adjustable parameters such as hammer hardness, velocity curves and the response to aftertouch.
In sound production, all these parameters are recombined to influence the final sound. Since only the original frequency profile is modified, this process is very rapid and can be obtained in real time with keyboards connected to modern computers. Also, there are much reduced requirements of disk space. My entire Pianoteq installation takes up only 52 MB, even though I have bought several additional harpsichords and pianos.
The sound is absolutely astounding. In my opinion, the Pianoteq simulations are far superior in quality to any built-in simulations that I've heard on any keyboard from major manufacturers (Casio, Yamaha, etc.). In fact, I am now a total Pianoteq addict. For my daily practicing I use a hammer-weighted MIDI keyboard connected with a midi cable to the Pianoteq simulation on my computer. For travelling, I have a lighter XKey keyboard that plugs into a small laptop.
And now comes the experiment. One of the wonderful things you can do with the Pianoteq simulations is to change between high-quality instruments at very low cost. Currently, Pianoteq offers three outstanding harpsichords and a clavicord simulation. The precise descriptions can be found on the Pianoteq site, and the photos below are from the same site.
Then for fair comparison, we also include two excellent harpsichords, each of which is found in larger symphonic offerings by Garritan and EastWest.
For final comparison, I also present a recording made with the Fluid R3 GM sound font.
I used a slow gigue by Giuseppe Domenico Scarlatti (Sonata Longo 106), so that one can hear the instruments' resonances clearly.
Quality of the reproduction
If you cannot hear much difference between these recordings, you may want to question the quality of your earphones.
Just the earphones, not any of the other parameters?
In a further set of experiments, I checked the long list of technical elements that could affect the sound quality of a recording of classical music: sampling frequency, band width, compression, the quality of sound reproduction and amplification chips, as well as the use of different earphones. It turned out that on computers built within the last ten years or so, a high-quality mp3 recording is generally adequate for classical music (192 kbps or higher). However, the earphones had the greatest influence on the sound quality. High-quality earphones permit much fuller sound with far less distortion than lower-quality earphones.
I happen to use Sennheiser and Audio-Technica earphones that cost around 200 Euros in 2016, and I'm much more satisfied with their reproduction than with earphones that cost considerably more a few years ago. Using these earphones, I can hear clear differences between all comparisons, including between the sound font recording (no. 5) and the various Pianoteq harpsichords.
The mp3 sounds below were created as 16 bit wave files and were compressed at the best mp3 setting of 320 kbps.
Which is the sound that you like best?
1. A simulation of a 1697 harpsichord made by Carlo Grimaldi of Sicily. This is a harpsichord for a princely court, built with the best woods and to the highest standards for a rich and clear tone. This is one of three Grimaldi harpsichords still in existence, and it was recorded in the Germanisches Nationalmuseum in Nürnberg, Germany. Produced with Pianoteq.
2. This is an exact 1624 copy of a harpsichord made by the famous Flanders builder Hans Ruckers II. Instruments by the Ruckers family were known at the time to be the best harpsichords in existence and were destined for very wealthy and princely buyers. Produced with Pianoteq, in the LB'+UB'+4 configuration.
Ruckers Harpsichord -- full
3. For the Ruckers harpsichord, it is useful to also present a different configuration that the Pianoteq engineers call "Dynamic".
Ruckers Harpsichord -- Dynamic
4. A harpsichord produced by François-Étienne Blanchet I. This instrument was produced in 1733 and was recorded at Frampton Cotterell, UK. An interesting detail is that it was decorated with monkeys making music or love, a depiction which probably found an excellent reception by the French royalties of the time. Produced with Pianoteq.
5. The Harpsichord KS included in the Personal Orchestra 5 by Garritan. No closer description is available.
6. The harpsichord included in the Symphonic Orchestra Gold by EastWest. No closer description is available.
7. An excellent copy produced by J.C. Neupert in 1941 of a clavicord used by Philipp Emanuel Bach, who was a strong advocate of clavicord use for teaching and private use. Chlavicords produce sounds by having small metal blades hit the string. They were affordable, easy to transport and quite popular from the 14th to the 19th centuries. Even today, there are many devoted clavicord players. In contrast to the somewhat muffled sound of other clavicords, this Neupert clavicord has a brighter and more present tone. At the same time, it must be remembered that the clavicord is capable of much less volume than a full-length harpsichord. Produced with Pianoteq.
8. For comparison: the harpsichord found in the Fluid R3 GM soundfont, by Frank Wen. This common soundfont was made available to the public domain in 2001 as an integrated product composed of various sets of samples that are copyrighted. A Google search has turned up no information on exactly how this soundfont was created or which type of harpsichord was modelled. The readme file suggests that the soundfont was carefully assembled and normalized from a variety of sources. Produced with Muse Score 2.
Fluid R3 GM sound font harpsichord
Fluid R3 GM (free sound font): https://musescore.org/en/handbook/soundfont