I'm still trying to replicate a piano with a synth (hobby project). So I plotted the spectrum of a single key (C4) to get an idea.
This is the plot of a sample of FL studio
This is how a Steinway & Son Model B looks like (forte)
The Steinway has way more features than the FL sample - this is why it sounds more interesting and organic.
(Description: Column B is the overall loudness and Column C, D, ... are the overtones 1, 2, ... normalized to the loudness - this is why they're not decreasing over time)
You can see the beat of the bichords/trichords which makes it harder to deduce a synth that can simulate the sound. I'm still unsure if higher frequencies decay faster over time. I'll try to increase the quality of the overtone plot.
The hardest part will be the attack phase, which contains noise and fast-changing spectra.
Once more the Steinway played on B0 (31 Hz) to make sure there's only a monochord in use. The first two harmonics are very weak and the 3rd harmonic is increasing! There seems to be some energy transfer.
Good luck. It is an indication of how complex an authentic piano sound is to synthesise that top manufacturers like Casio, Roland and especially Yamaha are still using massively multi-sampled concert pianos coupled with sophisticated systems of sample playback control (layer switching, tone filtering, decay transition, etc.) to create their digital models. Even physical modelling synthesis, successful with other acoustic string instruments like guitars, fails with the piano.
To synthesize an instrument I have to find the least complex setup that approximates the spectra close enough. Understanding all the physics is really helpful but not mandatory.
I'm currently looking for a paper that explains why the energy meanders through the harmonics in such a complex pattern - I've never heard of that phenomenon. Normally I'd expect an "attack"-spectrum that fades into "release"-spectrum (B0 seems to fulfill that). Sympathetic resonances are a good candidate.
Here's the Steinway spectrum again in linear space. Can anyone explain why the fundamental and overtone 2 (2*f) are symmetric?
Also for a beat I expected the pattern of the overtones to be "faster" according to their frequency (see my synthetic bichord example). But they seem all to be related to some global magic modulation.
Column B seems to be an indicator. For a bichord it has simple periodicity for a trichord it's more chaotic.
Now I have to find out:
- Why are overtone 1 and 2 exactly mirrored in the samples but not in the synthetic version (overtone 2 has a doubled frequency in it's amplitude for the latter)
- Why is the power graph sinusoidal in the samples but has a sharp peak in the synth version
Next up. Synthesize a realistically modelled cymbal crash