{"id":"0dceeab4-ec7e-4600-8f98-591e4c003ce7","arxiv_id":"2505.14448","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"Across 14 viola da gamba recordings by one performer, spectral frequencies mostly follow an exponential distribution and network centrality is correlated, which the authors attribute to the performer's style.","lead":"This paper models audio recordings of the bass viola da gamba as networks of note-to-note transitions and reports that most pieces fit an exponential distribution of spectral frequencies. The authors argue these patterns reflect the interpretive style of the single musician who performed all 14 recordings.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The network results cannot follow from the edge rule stated in Section 4.3: a graph with edges only between consecutive frequency components is a path, whose largest clique is at most 2, not 35–70.","rationale":"The Reader correctly identified the missing control and null model as a serious weakness, and also noticed the inconsistency between the described edge rule and the reported cliques. I agree with the rejection, but I would put the clique/edge-rule inconsistency first because it is an internal contradiction that independently sinks the network portion of the central claim, regardless of any future control experiments. The claim that network centrality and clique structure reveal the performer's interpretive style is not merely under-supported; it is not derivable from the stated algorithm. The statistical-distribution findings (exponential fits in 13 of 14 cases) might survive as an exploratory observation, but the abstract and Section 6 explicitly connect those regularities to the network construction, and that connection is the load-bearing part. The concrete test above is a single computational check that would settle whether the reported clique sizes are artifacts of an omitted method. Because the paper as written cannot reproduce its own network results, I concur with the REJECT verdict rather than a conditional acceptance.","tokens_in":12283,"tokens_out":2363,"duration_ms":26691,"concrete_test":"Reimplement the pipeline exactly as Section 4.3 is written for one audio file, e.g., '1Preludio re mayor': compute the FFT frequency sequence, map each frequency to its pair of nearest equal-tempered notes, and add only the two undirected edges per element (preceding–current and current–following). Report the number of nodes, the maximum degree, and the size of the largest clique. If the largest clique is 2 (or any value ≤2), Figure 5's 35–70 node cliques cannot come from the described method, which proves that an unstated edge rule or aggregation is required.","verdict_should_be":"REJECT","load_bearing_attack":"The central claim depends on both the statistical-distribution regularities and the network measures (degree-centrality correlations in Figure 4 and largest-clique sizes in Figure 5). The construction in Section 4.3 says: 'we can define a connection between two nodes based on the preceding and current frequency components, and other connection is formed by the current and following frequency components.' Taken literally, this creates only edges between consecutive elements of the frequency sequence, i.e., a path graph (or a subgraph of a path after duplicate-node merging). In an undirected path, every vertex has degree at most 2 and every clique has size at most 2. Yet Figure 5 reports largest cliques with 35 to 70 nodes, and Figure 4 reports substantial degree-centrality correlations across pieces. These results are impossible under the stated edge rule. An unstated aggregation step or a different edge rule (for example, connecting all pairs within some window, or building a complete graph per octave range) must have been used, but it is not described. Without that step, the network-based evidence for 'similar frequency fluctuations' and 'interpretive style' is not reproducible from the method as written. This is more load-bearing than the missing control group: even with a perfect multi-performer control, the network half of the claim would still rest on an undefined construction.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a method for characterizing a viola da gamba performer's interpretive style from audio recordings. For each of 14 live recordings by a single performer on a single instrument, the method computes an FFT spectrum, fits candidate statistical distributions to the resulting frequency components via Kolmogorov-Smirnov tests, and constructs an undirected 'network of sounds' whose nodes are pairs of adjacent equal-tempered notes and whose edges supposedly connect consecutive frequency components. The paper reports that the exponential distribution best fits 13 of 14 pieces, that degree-centrality correlations across pieces are mostly high, and that largest cliques contain 35–70 nodes. It interprets these results as evidence that the performer imprints a distinctive pattern of frequency fluctuations, and it states in Section 6 that 'We confirm that our hypothesis is correct.'","tokens_in":12486,"tokens_out":4005,"duration_ms":38651,"significance":"The topic is potentially interesting: connecting spectral statistics and network measures to performer identity could contribute to music information retrieval and performance analysis. The paper has some genuine exploratory virtues: it uses a real corpus of live recordings, integrates signal processing with music-theoretic note labels, and states an intention to share code and data. However, the central claim is not supported by the present analysis. The network construction as written is internally inconsistent with the reported clique sizes, the design lacks any control performer, instrument, or null model, and the statistical confirmation is circular because the same data are used to fit and to validate the distribution. These are load-bearing problems that prevent the results from being evaluated or reproduced. If properly reworked with a coherent network definition and a multi-performer comparison, the underlying idea could be worth exploring, but as submitted the significance of the claimed findings cannot be assessed.","major_comments":[{"comment":"The stated edge rule—connecting each frequency component to its immediate predecessor and successor in the sequence—produces a graph in which every edge lies between consecutive elements, i.e., a subgraph of a path. In such a graph every clique has size at most 2. The reported largest cliques of 35 to 70 nodes (Section 5, Figure 5) are therefore impossible under the described construction. The manuscript does not describe the additional aggregation step or alternative edge rule that would be needed to obtain these cliques, so the network results as reported are not reproducible and the network-based evidence for 'similar frequency fluctuations' cannot be evaluated.","section":"§4.3, Figure 5"},{"comment":"The central attribution of the observed regularities to the performer rests on the claim that 'the only constants in our musical cases were the same interpreter and his bass viola da gamba.' However, the design has no control performer, no control instrument, no null model of random frequency sequences, and no statistical comparison against recordings by other performers. Repertoire, recording venue, recording date, and recording equipment vary across the 14 files, so the same-performer attribution is confounded with these factors. The sentence 'We confirm that our hypothesis is correct' is therefore not supported by the presented analysis, which does not test the hypothesis against any alternative.","section":"§6"},{"comment":"The best-fit procedure estimates distribution parameters from each piece of music and then applies the Kolmogorov-Smirnov test to the same data, yet the reported p-values (e.g., 0.93 for '1Preludio re mayor') are interpreted as standard goodness-of-fit probabilities. No correction for estimated parameters (such as Lilliefors correction or a simulation-based null) is described. In addition, the candidate distribution list is taken from the authors' prior papers without external justification, and the exponential fit is then presented as evidence for the performer-style hypothesis even though no alternative hypothesis—such as repertoire or recording-chain effects—is tested. This circularity undermines the statistical-regularity claim.","section":"§4.2, Table 2"},{"comment":"Essential implementation details are missing: the audio sample rate, FFT window size and overlap, how the 'constituent frequencies' are selected from the spectrum (all bins, local maxima, or something else), and how the mapping to equal-tempered pitch-pair nodes handles repeated frequencies or frequencies outside the musical range. The statement that 'the database and the code will be available' on OSF gives no URL or release date, so the analysis cannot be replicated as submitted. These omissions are load-bearing because the network results, including the degree-centrality correlations and clique sizes, depend precisely on these undocumented choices.","section":"§4.1, §4.3, reproducibility"}],"minor_comments":[{"comment":"The spelling of the piece names is inconsistent: 'Arpegaita' appears in Table 1 while 'Arpeggiata' and 'Arpegiata' appear elsewhere; 'Sanite-Colombe' is a typo for 'Sainte-Colombe'. Please standardize the piece and composer names throughout.","section":"Table 1 / §3"},{"comment":"The exponential distribution is defined as f(x)=exp(-x), i.e., unit rate, but Table 2 lists two parameters for each exponential fit (e.g., 8.217e-05 and 29.797 for '1Preludio re mayor'). The parameterization used (scale, location, or rate) is not defined, making the reported fits ambiguous.","section":"§4.2 / Table 2"},{"comment":"Several references have errors: Massey (2020) should be Massey (1951); Bron and Kerbosch (2015) should be Bron and Kerbosch (1973); and there are formatting issues such as double commas in 'Cooley et. al.,,' and 'Hagberg et. al.,,'. Please check the reference list.","section":"References"},{"comment":"The text in Section 5 states that the degree correlation matrix uses Spearman correlation, but no details are given about which degree values are correlated (e.g., per-node degrees across pieces) or how ties are handled. Please clarify the computation and add a colorbar/legend to Figure 4.","section":"§5 / Figure 4"},{"comment":"The caption of Figure 3 states that only nodes belonging to the largest clique appear, but the text says the figure 'encapsulates the degree centrality and the largest clique in a network of sounds.' Please clarify what is plotted, and note whether the spiral layout is computed from the largest clique alone or from the full network.","section":"§4.4 / Figure 3"}],"recommendation":"reject","confidential_remarks":"The manuscript reads as a preliminary exploratory study. The most serious issue is the internally inconsistent network construction: the described edge rule cannot produce the reported clique sizes, which suggests that an unstated aggregation step exists. Even if that step were revealed, the central claim about performer identity would remain untested because only one performer and one instrument are used, with repertoire and recording conditions confounded. The statistical confirmation is also circular. These problems are not local presentation fixes; they require additional data (control performers/instruments) or a fundamental reframing of the claim, which is beyond the scope of a standard revision. I would encourage the authors to release the code and data, clarify the network construction, and consider a within-performer vs. between-performer comparison before resubmitting elsewhere."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Main take: the paper asks a good question—whether a performer's interpretive style leaves measurable statistical traces in raw audio—but as written the network half of the analysis cannot be right. Section 4.3 defines edges only between consecutive spectral components. That construction gives a path graph, where the largest clique is at most 2. Figure 5 reports largest cliques of 35 to 70 nodes. So some aggregation step or a different edge rule must be in play, and it is not described. This is not a minor gap; it makes the degree-correlation matrix and clique results unreproducible from the method as written.\n\nWhat is genuinely new and worth keeping: mapping each spectral component to a node defined by its two nearest equal-tempered notes is a clean discretization, and the degree-correlation matrix across pieces is a sensible exploratory instrument. The two recordings of the same Arpeggiata piece by the same performer (degree correlation 0.97, clique size 56 vs 57) provide a nice internal consistency check. The authors also explicitly flag the recording-versus-performance distinction, which is historically and methodologically careful.\n\nSoft spots, in order of severity:\n1. The graph-construction contradiction. The text needs a precise edge rule and either code or a small worked example before anything else.\n2. No control performer, no control instrument, and no null model. \"The only constants in our musical cases were the same interpreter and his bass viola da gamba\" is asserted, not demonstrated. Repertoire and the recording chain are equally constant.\n3. The KS p-values are computed after estimating parameters from the same data, so the p-values are invalid without Lilliefors or simulation.\n4. Data and code are promised but not available; the OSF link is not given.\n\nThe distribution-fitting result—exponential in 13 of 14 pieces—is a pattern, but without a null model it is unclear whether that is special to this performer or generic to spectral decompositions of harmonic music. The paper's own claim that \"we confirm that our hypothesis is correct\" overstates what an exploratory single-performer design can support.\n\nI would not publish this as is. But I would send it to referees rather than desk-reject it: the research question is legitimate, and the main flaws are fixable with a clearer construction, a control condition, and proper significance testing.","headline":"Interesting question, but the network construction as written cannot produce the reported cliques—the paper needs a major methodological fix before it is publishable.","tokens_in":13077,"tokens_out":4230,"would_cite":false,"duration_ms":45617,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Frequency fluctuations in recordings by one viola da gamba player collapse into a single exponential distribution in 13 of 14 pieces, and the paper attributes that shared signature to the performer's interpretive style.","keywords":["frequency fluctuations","complex network","bass viola da gamba","spectral decomposition","exponential distribution","degree centrality","clique","interpretive style"],"falsifier":"Record the same repertoire on the same instrument with several different viola da gamba players, or record the same player on a different viola da gamba, using the same microphone chain; if the exponential distribution and the high degree-correlation pattern appear in every case, the performer-specific interpretation is falsified. Alternatively, compute the best-fit distribution and largest cliques on shuffled or randomly relabeled spectral sequences; if the shuffled data reproduce the same regularity, the result is an artifact of the pipeline.","tokens_in":12014,"feed_emoji":"🎻","tokens_out":8786,"duration_ms":74604,"temperature":0.7,"pith_summary":"This paper sets out to show that interpretive style leaves a measurable fingerprint in the audio signal itself, not just in the score. The authors convert the spectral frequencies of 14 recordings by one bass viola da gamba player into a 'network of sounds,' then ask which statistical distribution best fits the frequency components and how the network's centrality and cliques behave. They find an exponential distribution in 13 of 14 pieces and strongly correlated degree-centrality patterns across most of the repertoire, despite the pieces spanning three composers, several recording dates, and three venues. Because the only constant across the recordings was the same interpreter playing the same instrument, the paper concludes that these large-scale regularities are the performer's imprint, with small-scale network differences capturing fine details of each performance. If the claim is right, listening to the spectrum of a recording could identify who is playing, independent of the written music.","feed_headline":"One viola player's recordings share one statistical signature","feed_subtitle":"14 pieces, three venues, one performer's exponential signature","key_machinery":"The argument is carried by the 'network of sounds': a graph in which each spectral frequency component becomes a node labeled by the pair of equal-tempered notes whose pitches bracket it (for example {E4, F4}), and edges connect each component to the component immediately before and after it in the spectral sequence. Degree centrality then ranks the most connected sound groups, and the largest clique, found with the Bron–Kerbosch algorithm, identifies functional groups of sounds that interact closely. The other load-bearing piece is the best-fit procedure, which uses the Kolmogorov–Smirnov test to choose among seven theoretical distributions; the exponential distribution wins in nearly every case, and this recurrence is what the paper treats as the signature of the interpreter.","core_discovery":"On the paper's own terms, the central discovery is that a diverse set of pieces played by one musician on one instrument shows a common statistical and network signature. After a fast Fourier transform of each audio file, the resulting frequency components fit an exponential distribution in 13 of 14 cases, with the single outlier ('4Otra cadencia') fitted by a gibrat distribution, a special case of the lognormal. When the frequency components are translated into a network of sounds—each node a pair of equal-tempered notes bracketing a frequency, each edge linking consecutive components—the degree centrality of nodes positively and strongly correlates across 78% of the pieces, and the largest cliques consistently span 35 to 70 nodes across the same low-to-middle note ranges. The two separate recordings of Abel's Arpeggiata, made months apart, show a degree correlation of 0.97 and similar clique sizes, which the authors read as evidence that the performer's frequency-fluctuation signature is stable while remaining sensitive to venue and recording equipment.","pith_inferences":["The same exponential fit might appear for any instrument with harmonic-rich, low-frequency-dominant timbre, so a control using a different musician on the same instrument, or the same musician on a different instrument, is needed to separate style from timbre and tuning drift.","Shuffling the spectral sequence or randomizing the equal-tempered labels would provide a null model; if the shuffled data still fit an exponential distribution and produce 35-to-70-node cliques, then those measures are not carrying performer-specific information.","If the method generalizes, it could serve as a forensic attribution tool for unidentified or historically disputed viola da gamba recordings, since the recurring distribution and network signature would be compared across candidate performers.","The use of three different recording venues across the dataset offers a natural stress test: the reported differences between the two Arpeggiata recordings hint that the method's fine-grained network measures may be sensitive to microphones and room acoustics, which future work could quantify directly."],"forward_implications":["A single performer's recorded audio should consistently show an exponential distribution of spectral frequencies and positively correlated degree-centrality measures across different composers, pieces, and recording venues.","Two recordings of the same piece by the same player will retain a high degree correlation (0.97 in the reported Arpeggiata pair) and similar largest-clique sizes, while differing in fine-grained node counts in particular note ranges.","Different musicians playing the same pieces should produce measurably different network-of-sounds fingerprints, making the method a candidate for algorithmic style comparison.","The exponential regularity coexists with small-scale network variations, which the paper interprets as the predictable and unpredictable components of the performer's style, consistent with the view that a performance is computationally irreducible."],"supporting_citations":[{"why":"Supplies the FFT algorithm that converts audio waveforms into frequency components for spectral decomposition.","marker":"Cooley et. al.,, 1965"},{"why":"The Kolmogorov–Smirnov goodness-of-fit test used to pick the best statistical distribution for each recording's frequency components.","marker":"Massey, 2020"},{"why":"Provides the statistical-regularities method for early music that this study transfers to a complex-network setting.","marker":"Lugo, I., and Alatriste-Contreras, M.G., 2025"},{"why":"Supplies the graph formalism (nodes and edges) and degree-centrality definition used for the network of sounds.","marker":"Albert and Barabasi, 2002"},{"why":"Implementation of degree centrality and clique finding on which the network measures are computed.","marker":"Hagberg et. al.,, 2008"},{"why":"Algorithm used to identify the largest clique in each network of sounds.","marker":"Bron and Kerbosch, 2015"},{"why":"Source for the Spearman correlation coefficient used in the degree correlation matrix.","marker":"Kendall and Stuart, 1973"}],"fun_headline_variants":["One viola player's music shares one exponential fingerprint","Frequency noise in bass viola tracks follows one musician's law","Exponential signature ties together 14 bass viola recordings","Bass viola player's frequency fluctuations show stable pattern","Same statistical signature in a viola da gamba performer's pieces"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The claim rests on the premise that the only thing shared by all 14 recordings—the same performer and the same instrument—is what produces the shared exponential distribution and correlated network measures; the study does not test any other performer, instrument, recording setup, or shuffled-frequency null model, so any of those could be the real source of the pattern.","fun_headline_variants_meta":{"raw":{"variants":["One viola player's music shares one exponential fingerprint","Frequency noise in bass viola tracks follows one musician's law","Exponential signature ties together 14 bass viola recordings","Bass viola player's frequency fluctuations show stable pattern","Same statistical signature in a viola da gamba performer's pieces"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000142,"raw_usage":{"total_tokens":1152,"prompt_tokens":912,"completion_tokens":240,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":528,"completion_tokens_details":{"reasoning_tokens":157}},"tokens_in":528,"tokens_out":240,"duration_ms":3727,"temperature":1.0,"reasoning_tokens":157,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T15:33:48.484013+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Record the same repertoire on the same instrument with several different viola da gamba players, or record the same player on a different viola da gamba, using the same microphone chain; if the exponential distribution and the high degree-correlation pattern appear in every case, the performer-specific interpretation is falsified. Alternatively, compute the best-fit distribution and largest cliques on shuffled or randomly relabeled spectral sequences; if the shuffled data reproduce the same regularity, the result is an artifact of the pipeline.","supporting_citations":[{"cited_title":"and Tukey, J.W","cited_arxiv_id":null,"evidence_quote":"Supplies the FFT algorithm that converts audio waveforms into frequency components for spectral decomposition."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The Kolmogorov–Smirnov goodness-of-fit test used to pick the best statistical distribution for each recording's frequency components."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the statistical-regularities method for early music that this study transfers to a complex-network setting."},{"cited_title":"and Barabási, A","cited_arxiv_id":null,"evidence_quote":"Supplies the graph formalism (nodes and edges) and degree-centrality definition used for the network of sounds."},{"cited_title":"and Swart, P.J","cited_arxiv_id":null,"evidence_quote":"Implementation of degree centrality and clique finding on which the network measures are computed."},{"cited_title":"and Kerbosch, J","cited_arxiv_id":null,"evidence_quote":"Algorithm used to identify the largest clique in each network of sounds."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source for the Spearman correlation coefficient used in the degree correlation matrix."}],"review_version":1}