Good article, when taken in context. The context for this article is precision analog signal processing at audio frequencies and the use of very small, high-capacitance capacitors. Some of this is relevant to guitar pedals, most of it is not. If you're designing a smart phone or sound card, then this article is very relevant.
Here's what TI doesn't say:
1) High-permittivity dielectrics like X5R are best used in switching power supplies and as power supply bypass capacitors. They were never intended for precision signal processing.
2) There are many considerations when choosing a coupling capacitor. Voltage coefficient is only one of them.
3) The -103dB noise floor in the graphs corresponds to a 17-bit A/D converter. It's the noise floor of the TLV320ADC5140 evaluation module.
The graphs all plot THD (total harmonic distortion) + N (noise) vs. frequency. The noise part of the signal is around -103dB w.r.t. a 1Vrms signal. 1Vrms = 2.83Vp-p for a sinewave. Db is a logarithmic scale. For voltage or current, every 20dB corresponds to a 10X change in amplitude. -103dB w.r.t. 1Vrms is 7uVrms or 20uVp-p. Changing the signal level has no effect on the noise, it only affects the distortion. THD = -60dB is equivalent to 0.1% THD. This is lower than any guitar amp or speaker cabinet at gig volumes.
Peccary is on the right track, but let me clarify a few things.
1) Total series impedance matters, not just the input impedance of the next stage. Some of the time, but not always, the input impedance of the next stage is the dominant factor.
2) Noise does not change with signal level. Signal levels inside a pedal can run as high as the power supply rails. For example, the maximum signal that comes out of an MXR Distortion+ might be around 400mVp-p, but the opamp inside is saturating and the signal level there is over 7Vp-p.
3) The higher MLCC values can have lower distortion for the same dielectric material. Still does not make them a good choice. Good luck finding a 47uF thru-hole MLCC that will fit on a pedal circuit board.
4) Film and tantalum are better for audio signal processing based on all of the factors. For us, the primary factors are size, leakage, noise, reliability and microphonics.
Parting shots:
Worrying about <1% distortion in a coupling capacitor when you're building a guitar pedal makes no sense to me. We deliberately make distortion in a dirt pedal. The noise and distortion of a PT2399 is off the scale on these graphs. The 1uF MLCCs that I bought from DigiKey to use in my Arachnid are X7S dielectric. According AVX: "X7S dielectric chip usage covers the broad spectrum of industrial applications where known changes in capacitance due to applied voltages are acceptable." A pretty poor choice if you're looking to squeeze the last drop of clarity from a high-end audio amplifier, but they sound just fine in the Arachnid. I used them them because that's what fit.