Source-linked AI summary
A computational framework for the acoustic characterization of Suikinkutsu (water harp cave)
Paul Haimes, Alarith Uhde, Daniel Moritz Marutschke
TL;DR
Suikinkutsu remain underdocumented computationally, and estimating their geometry is often difficult because reliable dimensions are unavailable. This study characterizes recordings from nine Kansai sites using complementary audio descriptors; the recordings retain broad timbral similarity while showing measurable differences, consistent with a bounded range of resonant behaviour. The findings are limited by the small recording sample, missing verified physical measurements, and uncontrolled field conditions.
Problem
Reliable dimensions for suikinkutsu are rarely available, while computational methods for documenting and comparing their acoustic characteristics remain underexplored.
Source: §1.2 · p. 3Method
The study extracts a literature-grounded combination of spectral and temporal descriptors from field recordings at nine Kansai-region sites to represent and compare their acoustic profiles.
Source: §2.1 · p. 3, §2.3 · p. 4Results
Pairwise mean-MFCC cosine similarities ranged from 0.850 to 0.999, while the descriptors also revealed measurable site-level differences within a bounded range of characteristic resonant behaviour.
Source: §3.1 · p. 5, §4 · p. 7Takeaways & Limitations
Multidimensional acoustic descriptors provide a way to document and compare suikinkutsu when physical measurements are inaccessible.
Source: §1.2 · p. 3The study has few recordings, lacks verified physical measurements, and uses uncontrolled on-site recordings where environmental variables and day-to-day differences may affect comparisons.
Source: §4.1 · pp. 7–8
Abstract
from arXiv · showhide
Suikinkutsu are partially concealed acoustic devices traditionally installed adjacent to stone water basins (tsukubai) in Japanese temple gardens. While previous research has examined their physical acoustics and cultural significance, computational methods for documenting and comparing their acoustic characteristics remain underexplored. This exploratory study introduces a computational framework for analyzing suikinkutsu using spectral and temporal audio features extracted from field recordings. Rather than estimating the physical geometry of individual suikinkutsu, the framework characterizes their acoustic profiles using a multidimensional set of computational descriptors, enabling systematic comparisons between recordings across different sites and installations. Audio recordings from nine suikinkutsu located at temples, shrines, and gardens in the Kansai region of Japan were analyzed using a feature-extraction pipeline that incorporates Fast Fourier Transform (FFT), Mel-Frequency Cepstral Coefficients (MFCCs), Root Mean Square (RMS) energy, spectral centroid, spectral bandwidth, and spectral contrast. The extracted features indicate that, although the suikinkutsu share broadly similar spectral and timbral characteristics, each exhibits a measurably distinct acoustic profile; a supplementary analysis of within-site and between-site variability suggests these profiles reflect a bounded range of characteristic resonant behaviour rather than a fixed acoustic signature. The proposed framework provides one of the first computational approaches for systematically characterizing and comparing suikinkutsu, contributing a practical foundation for future research and the digital preservation of this distinctive form of Japanese environmental design.
1 Introduction
Suikinkutsu are resonant water features with a long but underdocumented history, and prior work has chiefly examined their physical acoustics and cultural significance. This study proposes computational audio characterization as a way to compare and document them when physical measurements are difficult to obtain.
1 Introduction
- Suikinkutsu are partially concealed features beside stone basins, typically a buried pot where dripping water produces resonant echoes.A bamboo listening pole often invites visitors to hear the sound; some vessels are above ground.Source: §1 · p. 1, Figure 1 · p. 1
1.1 Literature Review
- Prior acoustic studies linked suikinkutsu sound to water-drop excitation and vessel-cavity resonance, while later work modeled these mechanisms computationally.Earlier work also examined cultural and garden-soundscape significance, but the proposed framework focuses on systematic acoustic comparison.Source: §1.1 · p. 2
1.2 Motivation for the Study
- Because recordings were made in situ under variable conditions, the study treats profiles as recording-level descriptions, not fixed or fully reproducible properties of each device.Environmental sound is present, but the study does not claim to isolate it with certainty or treat it as an object of study in its own right.Source: §1.2 · p. 3
- The study represents each suikinkutsu with computational audio descriptors to compare recordings when reliable physical dimensions are rarely available.Here, acoustic characterization means a multidimensional descriptor representation, rather than estimating vessel geometry.Source: §1.2 · p. 3
- The framework combines audio descriptors for digital-heritage documentation, extending systematic sound documentation to suikinkutsu not individually recognized in Japan’s 100 Soundscapes initiative.That initiative includes two named suikinkutsu among several hundred known sites nationwide.Source: §1.2 · p. 3
2 Method
The study recorded nine Kansai-region suikinkutsu and extracted complementary audio descriptors for comparison. It also compares repeated takes within sites against takes from different sites to assess the stability of site-level differences.
2.1 Data Collection
- Three usable takes were retained per site, trimmed to 10-second segments, yielding 30 seconds of analyzed audio per site across nine sites.The clearest take at each site was selected as its representative recording for site-level comparisons.Source: §2.1 · pp. 3–4
2.2 Audio Pre-Processing
- Each take was trimmed, peak-normalized, exported as mono WAV, and checked for clipping before feature extraction.The retained analysis segments were 10 seconds long, and peak levels were set to 0.0 dBFS.Source: §2.2 · p. 4, §2.1 · p. 4
2.3 Feature Extraction
- The feature pipeline combines standard spectral and temporal descriptors to form an acoustic feature vector for each recorded suikinkutsu.The descriptors were selected to reflect mechanisms such as cavity resonance and resonance decay; the combination was applied comparatively across field recordings.Source: §2.3 · p. 4
2.4 Within-Site and Between-Site Variability
- All three retained takes per site were compared pairwise as within-site or between-site pairs, with a permutation test assessing the difference in mean similarity.The analysis used 27 recordings, yielding 27 within-site and 324 between-site pairs; site labels were reassigned 10,000 times for the test.Source: §2.4 · p. 4
3 Results
Across the nine recordings, suikinkutsu show broadly similar timbral profiles alongside measurable site-level differences. Those differences reflect both resonant characteristics and recording conditions, so the profiles are better understood as bounded behavior than fixed acoustic fingerprints.
3.1 Timbral Similarity
- Pairwise MFCC cosine similarity ranged from 0.850 to 0.999; even the lowest pair remained comparatively high, while within-site takes were more similar.Within-site pairs averaged 0.99 similarity versus 0.95 between sites, a modest gap consistent with bounded characteristic behavior rather than fixed fingerprints.Source: §3.1 · p. 5, Figure 6 · p. 6
3.2 Acoustic Characterization
- The multidimensional descriptors distinguish site profiles, with Taizōin highest and Unryūin lowest in spectral centroid.Taizōin also has more moderate bandwidth than Miwa or Taimadera2, while Eikandō and Enkōji occupy intermediate positions.Source: §3.1 · p. 5, §3.2 · p. 6
3.3 Site-Specific Acoustic Divergences
- Site-level divergences reflect both suikinkutsu characteristics and recording conditions, while explanations for individual differences remain interpretive.Saijūji's manual pouring likely contributes to its distinct profile; ambient noise plausibly contributes to differences at Miwa and Kyoto Botanical Gardens, where low centroid and bandwidth also suggest vessel-related variation.Source: §3.3 · p. 6
3.4 Resonant Characteristics
- The recordings show resonant peaks across a broadly consistent frequency range, while RMS energy indicates variation in resonance persistence.Low-frequency activity at four sites is more plausibly environmental noise, but the boundary between noise and genuine resonance is uncertain; absent physical measurements also prevent attributing decay differences to a specific cause.Source: §3.4 · p. 7, Figure 7 · p. 7
4 Conclusion, Limitations, and Future Work
The nine-site exploratory analysis finds a bounded range of characteristic suikinkutsu resonance rather than a fixed acoustic signature, and offers a framework for documenting and comparing these structures. Limited recordings, unverified physical measurements, and in-situ conditions constrain the conclusions; future work could expand data collection and refine the analysis.
4 Conclusion, Limitations, and Future Work
- Suikinkutsu exhibit a bounded range of characteristic resonant behaviour, and the framework offers a practical method for documenting and comparing them as intangible sonic heritage.The recordings were collected at nine Kansai sites under variable environmental conditions.Source: §4 · p. 7
4.1 Limitations
- Limited recordings and unverified physical measurements constrain the study to acoustic characterization rather than precise geometric estimation.Source: §4.1 · p. 7
- Different visit days and repeated use of the same equipment and water-delivery method leave site differences partly entangled with environmental and recording conditions.The analysis cannot fully separate stable vessel identity from these sources of variation.Source: §4.1 · p. 8
- In-situ environmental noise and peak normalization complicate interpretation of low-frequency differences, and the analysis cannot establish a precise boundary between vessel resonance and ambient sound.The authors describe low-frequency attributions as informed interpretation rather than a rigorously validated threshold.Source: §4.1 · pp. 7–8
4.2 Future Work
- Onset-based drip-timing descriptors were excluded because individual impacts and sustained resonance could not be reliably distinguished across recordings.The difficulty was particularly evident where discrete dripping and continuous water flow both occurred.Source: §4.2 · p. 8
- Future work could validate onset detection against manual annotations, relate acoustic signatures to measured geometry, and build a crowd-sourced platform for larger datasets and additional analyses.Submitted audio would require manual quality review; similarity to verified recordings could also screen submissions.Source: §4.2 · p. 8