acceptodds
Withdrawn from ICLR 2027

Learning the Acoustic Background for Weak Vessel Observability through Temporal Probabilistic Deviations

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

Abstract

Rather than learning what vessels sound like, we model the acoustic background itself. This motivates a distinction between vessel presence and vessel observability: the former is a geometric condition, whereas the latter depends on whether the received acoustic signal is distinguishable from the prevailing soundscape. We introduce Probabilistic Background Support Modelling (PBSM), a background-first framework that learns the local operational acoustic background from no-contact recordings and identifies statistically unusual 30-second observations as acoustic departures. Rather than assigning these departures to a predefined acoustic class, PBSM reduces large passive-acoustic datasets to short-duration observations warranting further investigation. An Empirical Background Anomaly Percentile (EBAP) expresses departure relative to the learned background, while a matched 30-second no-contact control accounts for departures caused by temporal representation. Acoustic scoring is blind to vessel position, range, Closest Point of Approach (CPA), and identity. Identified departures are subsequently examined for substantive spectral excess, with AIS-derived vessel geometry introduced only afterward as an independent physical reference. Across eight station-season datasets, short-duration background variability can itself produce substantial acoustic departure, demonstrating that departure alone cannot be interpreted as vessel detection. The prevalence and discriminative value of PBSM departures vary across operational environments, while spectral characterization can retain meaningful acoustic structure even where background departure is prevalent. Where PBSM departure, substantive spectral excess, and vessel proximity converge, increasingly strong acoustic evidence is observed as CPA decreases. The principal contribution is their integration into a background-first evidence chain for acoustic observability: learn the local operational background, identify unusual observations blindly, characterize their spectral structure, and use independent vessel geometry to interpret the resulting evidence.

Primary area:
probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)
Keywords:
passive acoustic monitoring, underwater acoustics, probabilistic background modelling, acoustic anomaly detection, vessel acoustic observability, Gaussian mixture models, empirical background anomaly percentile, acoustic signal processing

Reviews

Among:
Shape:
1234this paper, Bottom 1%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.00 over 3 reviews, above 0% of the others.
Reviewer_CaDzOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

This paper proposes PBSM, a background-modeling approach for passive acoustic vessel monitoring. Instead of learning vessel signatures, it fits a density model of the acoustic background using "no-contact" recordings (no qualifying AIS vessel present). Each approximately 11-minute no-contact recording is represented by a Welch log-PSD (0-1000 Hz), reduced to 64 PCA components and modeled with a 12-component full-covariance GMM, separately for each of eight station-seasons (4 stations, 2 seasons). One-contact recordings (exactly one AIS vessel) are split into 30-second windows, scored by GMM log-likelihood, and converted to an empirical percentile (EBAP) relative to the background scores. Windows with EBAP ≤\le 5% are labeled "acoustic departures". Since the model is trained on full-length PSDs but tested on 30-second windows, held-out no-contact recordings split into 30-second windows are used as a matched control. Departures are then checked for spectral excess (at least 10 bins above the median background by 3 dB or 6 dB), and AIS-derived CPA is used after scoring to see whether departures and spectral excess change with vessel distance.

The main findings are as follow: (i) the 30-second representation alone produces very high departure rates in the control (16-100% of windows, 25-100% of recordings), so departure by itself is not detection; (ii) the enrichment of one-contact over the matched control is about 1.0×1.0\times in 5 of 8 station-seasons and at most 1.94×1.94\times (Station 5, 2016-17), and; (iii) at Station 2 (2015-16), the departure rate and the 6 dB spectral excess decrease with CPA, and spectral excess is more frequent among one-contact departures than among control departures.

Critical strengths
  1. The matched 30-second no-contact control is carefully constructed. It shows that most "departures" come from the mismatch between full-duration training PSDs and 30-second test windows, not from vessels (Table 1: 99-100% of control windows depart at Station 5 (2015-16), Station 6 and Station 20). Reporting this avoids an inflated detection claim, and it is a useful cautionary result for anyone trying to apply background or OOD models to operational PAM data.

  2. The evaluation does use real data from eight station-season pairs with a fixed protocol and no curation or per-site tuning. The change at Station 5 between seasons is a particularly interesting finding on how detectability depends on deployment conditions.

Critical weaknesses

There exists little to no quantitative detection evaluation nor baseline, so the value from the method cannot be accurately assessed.

The results are only prevalence rates at a single operation point. Paper reports no (AU)ROC, precision-recall, detection vs. false alarm analysis. There are also no comparisons (even simple ones like a broadband or band-limited energy detector against a running background, a density model fitted on 30-second no-contact windows (to account for the representation mismatch), standard OOD scores (k-NN distance, Mahalanobis, isolation forest, AE reconstruction), or an existing supervised ship-noise detector, etc.). This especially matters because a simple energy detector may perform just as well. The spectral-excess criterion is essentially a broadband level test: the "strong" examples in Figures 16 and 18 exceed the background by at least 6 dB in all 257 bins. At Station 2, where PBSM departure is in fact informative, the effect is small, not to mention no information carriage in the majority of station-season pairs.

2. Several design choices effectively undermine the "blind" or "independent" claim and the statistical interpretation.

(a) Section 3.3 states that PCA/GMM configurations were trialled "with their behaviour examined relative to independently derived vessel CPA" before freezing. Model selection therefore used vessel geometry, so the CPA analysis is NOT an independent validation. It is also unclear which station-seasons were used for this, and whether they overlap with the reported results.

(b) The train/test mismatch in representation is a design choice so does not constitute a property of the problem. Fitting the background on 30-second no-contact windows (with recording-level splits) is one alternative, and the paper does not explain why it was not used (or not reported).

(c) A recording counts as a departure if any of its roughly 22 dependent windows falls below 5%, so recording-level rates are inflated by multiple testing. The paper itself notes that EBAP is uncalibrated and that the windows are dependent.

(d) Low GMM likelihood is two-sided. A window can depart because it is quieter or spectrally different from the background (e.g., Figure 17: EBAP 0.000% at 0.8 km CPA, maximum excess 1.8 dB), which makes it a questionable proxy for vessel observability.

(e) Basic statistics are nowhere to be found. The number of no-contact and one-contact recordings, windows and unique vessels per station-season, the sample size per CPA bin, the sampling rate, and any confidence intervals or tests are simply not found. Without these, it is impossible to tell whether differences are actually meaningful.

Overall, the ML component (PCA, GMM likelihood and a percentile threshold) is a basic standard and close to prior background-GMM work, as the authors acknowledge. Given the limited methodological novelty and the lack of any comparative evaluation, the paper would fit better in an underwater acoustics or ocean engineering venue.

Questions for authors
  1. Could you report window-level and recording-level ROC/PR curves (or AUROC) for one-contact vs. held-out no-contact for each station-season, together with at least (i) a broadband/band-level energy detector and (ii) a GMM fitted on 30-second no-contact windows? If PBSM clearly outperforms, I would reconsider my assessment.

  2. Which station-seasons and data were used when trialling the PCA/GMM configurations against CPA? Were the held-out no-contact and one-contact sets involved?

  3. Please give the number of recordings (no-contact train/held-out, one-contact), 30-second windows and unique MMSIs per station-season and per CPA bin, with confidence intervals for the percentages in Table 1 and Section 4.2.

  4. How many of the windows with "substantive spectral excess" are broadband level shifts (most or all of the 257 bins above threshold) rather than narrowband or tonal components? How does the spectral-excess criterion perform on its own as a detector without PBSM?

  5. How are vessels without AIS (small craft, etc.) handled in the no-contact set? How was the "qualifying" contact radius chosen so that no-contact recordings are actually free of audible traffic?

  6. Are the CPA trends in Section 4.2 for Station 2 (2015-16) only? Do they hold after controlling for vessel type/size and speed?

Suggestions: calibrating EBAP on a separate calibration split (as mentioned by the authors) would allow the 5% threshold to be read as a false-alarm rate. The Station 5 seasonal contrast would also be more informative with the environmental and deployment covariates proposed in the Discussion.

Reviewer_HBUrOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

This paper studies the problem of identifying acoustic signatures of vessel sounds, by proposing a new approached called Probabilistic Background Support Modeling (PBSM). The main methodology is a GMM density estimation. The method is evaluated on a dataset with performance shown but no baseline method compared against. The main contribution of the paper is presented as the integration of multiple steps of the identification pipeline into an "evidence chain for acoustic observability".

Critical strengths

None

Critical weaknesses
  • The evaluation section does not compare against any baseline method, but simply presents the proposed method's performance. This severely undermines the validity of results of this paper. The evaluation metric itself is lacking too, as there's no presentation of ROC or PR curves.
  • A closely related work (Mihailov, 2025) is mentioned in Related Work but never evaluated upon. It is unclear if this paper has any methodological contribution beyond this related work.
  • There is in general little to no methodological innovation in this paper, and it's not clear to what extent this paper improves upon the results in the relevant area. Overall this is not expected of the quality of an ICLR paper.
Questions for authors
  • Can you compare your proposed methods with some baseline methods commonly used in literature for this problem/area?
  • Can you add standard reporting metrics for detection problems, such as ROC curves or PR curves?
Reviewer_4t2nOpenReview ↗
Rating
1
Confidence
2
Informativeness
2
Presentation
2
Soundness
3
Read review
Paper summary

This paper proposes Probabilistic Background Support Modelling (PBSM) for identifying unusual observations in passive underwater acoustic recordings. The method learns a local background distribution from no-contact recordings using power spectral density features, PCA, and a Gaussian mixture model. It then scores temporal windows using an Empirical Background Anomaly Percentile (EBAP), characterizes their spectral excess, and examines their relationship. Experiments across eight station–season datasets show that short-duration background variability can itself produce substantial acoustic departure, highlighting the distinction between vessel presence, acoustic anomaly, and acoustic observability.

Critical strengths
  1. Useful temporal-representation control. The matched 30-second held-out no-contact population reveals an important consequence of training on full-duration recordings while evaluating short windows.

  2. Transparent reporting across environments. The evaluation includes saturated and less informative station–season results. Splitting at the original recording level also prevents windows from the same recording from appearing in both training and held-out sets.

Critical weaknesses
  1. The incremental value of PBSM screening remains unclear. In Table 1, both no-contact and one-contact recording-level departure prevalence reach 100% in four station–season datasets. The matched control helps interpret this saturation but does not restore screening selectivity. The paper does not compare against a background model trained on 30-second windows or against spectral-excess analysis without PBSM filtering.

  2. The methodological novelty primarily lies in integration. PCA, GMM density estimation, and empirical percentiles are established techniques. The cited work by Mihailov [1] already uses a background-only GMM, log-likelihood scoring, and a lower-percentile anomaly threshold. EBAP improves interpretability but does not introduce additional discriminative information beyond the underlying score. The integrated framework could still be valuable, but its contribution requires clearer evidence of incremental benefit.

  3. The independence of the CPA-based evaluation is insufficiently documented. Section 3.3 states that PCA and GMM configurations were examined in relation to vessel CPA during development. Although CPA is excluded from acoustic scoring, the paper does not clearly establish whether configuration-development data were separated from the final CPA analysis.

  4. Figures 1 and 2 contain too little information to adequately represent the content referred to in the caption.

[1] Characterization and automated classification of underwater acoustic environments in the western black sea using machine learning techniques

Questions for authors
  1. Under the same recording-level split and candidate-review budget, how does PBSM compare with a model trained on 30-second background windows and with direct spectral-excess screening? What additional evidence or reduction in review workload does PBSM provide?

  2. Which components contribute measurable improvements over background-GMM scoring with a percentile threshold? Can targeted component comparisons establish the benefit of the integrated pipeline?

  3. Were the recordings used to select PCA/GMM configurations completely separate from the final evaluation? Please specify the development and evaluation partitions.

  4. As an incremental approach, its unique contributions should be articulated more clearly. Additionally, it is recommended to include comparisons with some recent papers from top-tier conferences.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.