Acoustic Detection of Spaceflight Fluid Shift: A Consistent In-Flight Decline in Harmonics-to-Noise Ratio Across Three ISS Crew Members, with an Operational Model for Voice-Based Crew Health Monitoring

Abstract Cephalad fluid redistribution is among the earliest and most consistent physiological consequences of orbital insertion, yet it is difficult to monitor in flight: the measurements that resolve it require consumables, hardware mass, or crew time that exploration-class missions cannot spare. Voice is a candidate readout — phonation depends on vocal-fold surface hydration and viscoelasticity, both of which fluid shift perturbs — but no controlled in-orbit vocal corpus has previously existed against which to test it. We report such a corpus and the result of that test. Three International Space Station crew members performed a fixed vocal battery on a certified crew tablet, twice daily, across a pre-flight phase and mission days 3–14, yielding 44 sessions and 254 verified recordings with a pre-flight baseline acquired through an identical acquisition path. Every recording was retained in two parallel forms — the raw acquisition and the same material after a noise-suppression stage — and both were measured with a single Praat/Parselmouth pipeline. Harmonics-to-noise ratio (HNR) declined between the pre-flight and in-flight phases in seven of the eight testable crew × vowel cells, median − 2.11 dB. At the recording level, with all three vowels pooled and each crew member's own pre-flight median subtracted, the decline is − 2.70 dB (23 pre-flight vs 96 in-flight recordings, p = 0.019), reaching significance within a single crew member (− 5.53 dB, p = 0.011, rank-biserial − 0.72); restricted to the two vowels sampled completely in every crew member it is 6 of 6 cells at − 2.47 dB ( p = 0.0038). The effect is specific to HNR: shimmer and CPPS, which share its sensitivity to added acquisition noise, decline in only 2 and 3 of 8 cells respectively, so a drift in recording conditions between phases — which would have displaced all three together — does not explain it. A reduction in HNR is an increase in turbulent, aperiodic energy during phonation, the direction predicted by laryngeal mucosal fluid loading. The change presents as a sustained step between phases rather than a progressive in-flight trend, consistent with fluid shift reaching a new steady state within hours of insertion. Because acoustic measurements are vulnerable to their acquisition chain, we established what that chain contributes before interpreting the result. Retaining both processing versions allowed this to be measured directly: the noise-suppression stage moves shimmer by − 3.35% and HNR by + 3.56 dB (paired, n = 63, p ≈ 10⁻¹¹), while leaving F₀ and CPPS unchanged (CPPS p = 0.94). The phase contrast above is computed within a single processing version, so this term cancels; quantifying it is what licenses the physiological reading rather than undermining it. The same contrast applied to 135 voluntary cough recordings gives no consistent direction in any measure, which we report as evidence for task selection: sustained phonation is self-normalising, whereas cough acoustics depend on applied effort and instantaneous mouth-to-microphone geometry. We report two further chain properties for reuse: cough timing depends 9.90-fold on segmentation convention, and within-session dimensionless ratios are stable across processing versions (within-crew CV 11.3–15.0%). The operational case is quantified: 18.2 s of crew recording time, zero consumables, zero incremental mass, and 624 B of downlink per session against 568 KiB of raw audio — a 932-fold reduction with on-board measurement. We give design requirements for follow-on missions in directly usable form: nine sessions per phase per crew member reach d = 1.5 at 80% power, against d ≥ 1.91 permitted here, and pre-flight sessions cost nothing in flight resources. No blood or imaging reference was obtained, so the fluid-shift attribution is an interpretation consistent with the observed direction and specificity rather than a demonstrated mechanism.