This paper presents Infra-Net, a parallel dual-branch architecture that integrates log-scattering wavelet transforms with a confidence-based fusion mechanism for infrasound classification. The approach is methodologically sound and physically motivated, with the column-wise multi-view feature treatment demonstrating a notable degree of novelty. The results on the public dataset are impressive; however, the significant accuracy drop observed on real-world CTBTO data raises concerns regarding model generalization and overfitting. The manuscript would benefit from a more rigorous ablation study and a detailed analysis of computational complexity. Therefore, author should make major revisions before considering potential publication in this journal.

Specific comments are as below:

1. Please clarify the novelty of "multi-view" input, authors should explicitly contrast this with prior work that concatenates features, and justify why treating columns independently is theoretically superior rather than an arbitrary data expansion.
2. Page 21, Line 539: Please provide statistical significance testing for the CTBTO results. The reported 82.07% accuracy on CTBTO data is only marginally higher than simpler architectures (e.g., Mc at 81.89%).
3. Authors should discuss how to address the substantial generalization gap between datasets, because the performance drop from 100% (LOTIS) to 82.07% (CTBTO) suggests potential overfitting to the LOTIS acoustic environment.
4. In Table 6, please clarify how the 100% accuracy on LOTIS was computed—was this the average across folds, the best fold, or a separate hold-out test? The confusion matrices in Fig. 10 do not reflect the fold-averaging described.
5. Page 13, Line 325: Please provide visualization of the confidence-based fusion process with a supplementary figure showing how the per-column probabilities and the final inner product vary for a correctly classified vs. misclassified CTBTO sample.
6. Suggest revising the final paragraph of the current Conclusion to include a brief statement on limitations and future domain adaptation work.

The adjective “robust,” used by the authors in the title, encapsulates the quality of the study and clearly and unambiguously reflects the content of the article. The detection of infrasound associated with geophysical events such as earthquakes, volcanic eruptions, or gravitational movements is well-established and applied both in scientific contexts—such as the USGS’s monitoring of volcanic activity or the CNR in Italy, inspired by Giovanni Gregori—and in professional settings for monitoring gravitational phenomena, particularly during major construction projects and infrastructure development. The manuscript addresses scientific and technical issues of significance both within the field of NHESS and within Earth Sciences. The creation of a network to distinguish the type of infrasound produced by natural events from that of anthropogenic origin represents a novel element in this area of research, as does the use of new technological devices compliant with international standards. The methodology adopted for conducting the research is effectively described, and the scientific approach is sound, as evidenced by the results that support the robustness of the technique used and by the study’s conclusions. The description of the experimentation is sufficient to stimulate the interest of other research centers to pursue similar work. Both the title and the abstract are aimed at a specialized audience, given their technical nature.

The size, quality, and legibility of each figure are not always appropriate for the type and amount of data presented. The authors give proper credit to previous or related work while specifying their own contribution. In this context, the number and quality of the references span a period of about thirty years and are appropriate and accessible for use by other scientists. The presentation of the study is well-structured but intended for scientists in the field, while the length of the paper, though substantial, is appropriate for an international scientific context, thanks in part to precise technical language and high-quality English. The document does not include supplementary material.

The authors could specify whether monitoring is continuous and active 24/7, and add a few words about the acronyms that, while clear to researchers in the field, could help non-specialist researchers—such as the acronym CTBTO. In lines 68, 70, and 72, there are three capital letters after the colons, and Figure 1 could use a bit more explanation. Furthermore, Figure 5, although illustrative, is not entirely clear. In line 314, on page 12, the meaning of “neurons” could be clarified.