One ladder: from the thermal floor of air to a dog barking in a living room

Voice is recorded in anechoic chambers, in sound-treated booths, in consulting rooms, aboard spacecraft, and on whatever device happened to be nearby. These are usually treated as different fields with different standards. They are one ladder with a floor set by thermodynamics, and this paper puts every rung on it in a single currency. THE SPACEFLIGHT STEP, MEASURED. From the leading silence and the phonation of 231 recordings made by three crew members before and during a spaceflight, the signal-to-noise ratio at the microphone falls from a median of 28.6 dB in the pre-flight quarantine room to 16.7 dB in orbit. That is a ratio and carries no calibration assumption. Passed through the field's own measured response curve it is 7.30 dB of harmonics-to-noise ratio lost to the recording alone, which at a measured 0.0427 percentage points per decibel manufactures 0.312 percentage points of jitter and shifts cepstral peak prominence by 5.62 dB -- before the larynx has done anything. THREE QUARTERS OF AN OBSERVED CHANGE. The same crew member's observed pre-flight to in-flight change in cepstral peak prominence was -7.44 dB. The recording step accounts for -5.62 dB of it, 76 per cent. For jitter the share is 22 per cent. This is the same authors quantifying their own secondary comparison, and the conclusion is that it cannot support a physiological reading. BELOW THE SPACECRAFT SITS THE LIVING ROOM. In 440 laboratory recordings of 21 cats the median harmonics-to-noise ratio is 15.67 dB; in 77,194 social-media recordings of 156 dogs it is 7.87 dB. Spaceflight recordings sit BETWEEN them, at 10.4 to 13.8 dB by vowel. A dog filmed on a phone is a worse acoustic environment than the International Space Station. The gap between the two animal corpora, 7.80 dB, is larger than the 6.01 dB separating a sound-treated booth from the worst clinical room in the published literature. ONE STATISTIC TRAVELS THE WHOLE LADDER. The ratio of between-individual to within-individual variance says whether a parameter carries identity or state. In the controlled cat corpus fundamental frequency carries identity (2.10); in the uncontrolled dog corpus it does not (0.36), while the three parameters that never carried identity are unchanged. Because those corpora differ in species as well as discipline, the same statistic was computed on synthetic material where the truth is fixed by construction: across benign recording conditions the ratio for fundamental frequency is 102.9, and across hostile ones, with the truth unmoved, it is 0.608. WHETHER A VOICE PARAMETER IS AN IDENTITY SIGNATURE OR A STATE INDICATOR IS A PROPERTY OF THE RECORDING DISCIPLINE, NOT OF THE PARAMETER. Any claim that a feature is a stable biometric of an individual is a claim about the corpus it was measured in. The analysis script is included. The two animal corpora are non-commercially licensed and are not redistributed, and neither are the feature tables derived from them; the extraction code is included instead, so that anyone holding the corpora under their own terms can regenerate them exactly. No audio is shared.