While we wait for further developments on this topic, a post on the motherland has become an inspiration. This is a concept that I will work over the next few days. Can we create a sonic fingerprint of Uruka’s viola technique and use it to independently verify the dox?
Notes created by a musical instrument are vibrations in the form of a fundamental frequency waveform and a series of harmonics that can be analyzed by using frequency spectrums. The type of instrument and the technique used to play it cause subtle modifications to the shapes of the frequency spectrum for a given note. There are three obvious pieces of the spectrum to examine. The first is the attack, which is the first part of the note when the waves are excited. The second is the decay, which is how the waves are damped and relax back to their unexcited state. The third is the harmonic proportion, which is the magnitude of the harmonic waves compared to the fundamental frequency.
Viola Acoustics
There is some baseline data on viola acoustics to use as a reference, as seen in the link provided plus the image of harmonic proportions taken from a second reference. We’ll know if what we get makes sense.
There is a fourth pattern to examine, which is the timing of the individual notes against the nominal beats per measure of the time signature.
The viola is a complex instrument to play with at least two dozen different bowing techniques, from detache to spiccato, and requiring precise finger technique. Variation in the bowing technique should impact the attack the most, though there will also be some impact on decay. Finger technique should impact decay considerably, especially if fingers are used to provide vibrato.
These two videos of viola playing were posted on the motherland. Unfortunately, the viola is lost in the midrange noise from the other instruments and is impossible to extract. However, what you see in the videos is a fairly aggressive bowing technique and a unique spastic finger movement to apply vibrato. Hopefully, that will provide enough modulation to create a unique acoustic fingerprint. We just need a good sample.
Dox Sample #1
Dox Sample #2
Further searching yields this video, which in the third movement provides a solo with heavy vibrato and in the fourth movement provides a solo with rapid bowing and no vibrato. This should be sufficient to examine the attack, decay, and harmonic proportions under two separate styles of playing. We can also look for patterns in timing in both a slow and fast piece.
Dox Sample #3
As an aside she’s quite good. Watch her play with her eyes closed with an expression of pure joy in the fourth movement.
Uruka has done several viola streams. I selected one with several different pieces of interest. There are two vibrato heavy pieces at approximately 25:00 and 54:00, and an uptempo piece at 1:10:00.
Uruka Sample
There are potential problems in making this comparison. There is background music, though the viola is prominent and can probably be extracted. Uruka turned on an effect which is going to skew the frequency spectrum. We don’t know how the viola is miked in any of these cases. We don’t know the microphone response curves, where relative to the rejection lobes the instrument is positioned, input levels, or any of that. Unfortunately, all of the samples are like this. We’ll have to hope it is miked well enough in all of the samples such that the contributions from those factors are not significant.
We also need a control group of known different violists to confirm that any acoustic fingerprint we can find is different from other violists. I found a set of different viola solos by various musicians that should cover various playing styles and recording setups.
Control Sample #1
Control Sample #2
Control Sample #3
Control Sample #4
Now that we have a data set for our analysis we can do a few things.
We can extract sequences of notes and align them against click tracks corresponding to the time signatures of the pieces. Do the dox sample and the Uruka sample show consistencies in timing? Are they both consistently early or late by the same amount compared to the click tracks? Do they have rushed or delayed notes under similar circumstances?
We can extract individual notes from the samples and run them through a fast Fourier transform to generate frequency spectra. We can compare differences in the shapes of the spectra among the controls, the dox, and the Uruka sample to try to identify telltale similarities. Enough similarities in common between the dox and Uruka sample but missing from the controls will serve as a fingerprint to confirm the dox and Uruka playing styles are identical.
We already have independent confirmation that the dox is accurate because of Urudog and other corroborations. This will be more of a validation of the method. Will it work? Let’s find out.