{"id":"9f692c21-7dd6-4d34-82d5-6c42609fc614","arxiv_id":"2607.19023","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Drivers detect smaller speed changes with internal combustion engine interior noise than with electric vehicle or silent sound at 100 km/h, with engine-order tonality proposed as the cue.","lead":"This study measured how well drivers can tell speed differences from watching video while hearing interior car noise. At highway speed, engine noise allowed finer speed discrimination than electric motor noise or silence, suggesting EV sound design should add speed-linked tonal cues rather than just volume.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reported ICEV JND at 100 km/h (1.93 km/h) lies below the stated 2 km/h minimum staircase step, so the headline ICEV advantage may be a quantization-floor artifact rather than a resolved threshold.","rationale":"The reader's weakest assumption was the loudness confound; that is a real open question, but the staircase-floor issue is sharper because it is an internal inconsistency in the paper's own numbers. The paper's strongest claim is the 100 km/h JND ranking ICEV < EV < Silence. If the ICEV estimates are quantized at 2 km/h, the exact reported means cannot be trusted and the size of the ICEV advantage is unknown; the tonality explanation built on the 0.49% frequency shift corresponding to 1.93 km/h would need to be re-derived from whatever below-floor thresholds are actually resolvable. The 40 km/h data and the silence-vs-sound contrast are less affected, so the paper is not wholly invalid. The right disposition remains conditional: the experimental design needs either a finer staircase or an explicit report of floor-hitting rates, and the raw-data check can settle it quickly. Because the reader's stated weakest assumption and my strongest concern are different but both target the same 100 km/h ICEV comparison, agreement is partial.","tokens_in":9267,"tokens_out":8132,"duration_ms":74722,"concrete_test":"Download the Zenodo/GitHub raw data and, for each 100 km/h ICEV staircase, list the comparison-speed offsets at the retained reversals. Check whether any value below 2 km/h occurs. (1) If all values are even integers, recompute the JND means; the reported 1.93 must change or the stated minimum step is wrong. (2) If some values are below 2 km/h, count how many participants are at the 2 km/h floor; then re-run a small validation sample (e.g., 10 participants) with a 1 km/h or 0.5 km/h minimum step. If the ICEV mean decreases further while EV and Silence are unaffected, the ICEV 'advantage' is partly an artifact of hitting the staircase floor; if it does not move, the floor was not binding and the original comparison stands.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The most load-bearing problem is an internal arithmetic inconsistency in the staircase procedure. In 'Procedure' the authors state that the minimum step size at 100 km/h is 2 km/h, and in 'Data Analysis' the JND is computed as the arithmetic mean of reversal levels. With a 2 km/h grid, every reversal level must be an even integer offset (e.g., 2, 4, 6 km/h), so no participant-level JND can be below 2 km/h, and the across-participant mean cannot be 1.93 km/h. Table 2 reports the ICEV JND at 100 km/h as 1.93 km/h. This means either the minimum step was actually finer than stated, or many ICEV staircases sat at the 2 km/h floor and the exact 1.93 value is not reproducible from the stated algorithm. Because the central empirical claim is that ICEV yields a lower JND than EV at 100 km/h (1.93 vs 3.48 km/h), a floor-bound ICEV distribution undermines the quantitative contrast: the true ICEV threshold may be at or below 2 km/h, and the magnitude of the advantage is unresolved. This is a more direct threat to the central result than the loudness confound, and it should be settled from raw staircase data before interpreting the tonality mechanism.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports a within-subject psychophysical experiment measuring just-noticeable differences (JNDs) in speed under three interior-sound conditions (ICEV, EV, silence) at reference speeds of 40 and 100 km/h. Thirty participants completed a 2AFC task with visual driving clips accompanied by auralized vehicle interior noises. The authors report that at 100 km/h, the ICEV sound produced a significantly lower speed JND (1.93 km/h) than the EV sound (3.48 km/h) and silence (5.15 km/h), while at 40 km/h the ICEV and EV conditions did not differ. They interpret the highway-speed difference as evidence that audible engine-order tonal components whose frequencies shift with speed provide a trackable spectral cue, and they recommend that EV sound design introduce speed-contingent tonal components rather than simply increasing loudness. The paper includes open-source code and archived stimuli.","tokens_in":9544,"tokens_out":8874,"duration_ms":80588,"significance":"If the result holds, the paper makes a useful contribution to an underexplored question: whether interior sound affects not just speed estimation bias but the perceptual sensitivity to speed changes. The within-subject design, use of realistic auralized stimuli, and public availability of code and data are strengths, and the practical design implication for EV sound engineering is potentially valuable. However, the central claim that the ICEV advantage at 100 km/h is specifically due to engine-order tonality is currently supported only by post-hoc psychoacoustic analysis; the experimental design does not independently manipulate tonality, and two confounds (absolute sound level and staircase quantization) leave the quantitative result open to alternative explanations. The observed effect may be real, but the manuscript's interpretation and the precision of the headline JND values require substantiation before the conclusions can be accepted.","major_comments":[{"comment":"The reported ICEV JND of 1.93 km/h at 100 km/h is at the resolution limit of the adaptive staircase, which had a minimum adjustment step of 2 km/h. If a zero offset was never presented, then every reversal level is at least 2 km/h and the mean of reversal levels cannot be below 2, making 1.93 inconsistent with the stated procedure. If a zero offset was allowed, then the threshold estimate lies within one step of the smallest nonzero stimulus, meaning the staircase has not resolved the threshold below the 2 km/h bin; the true ICEV JND could be anywhere from near 0 to 2 km/h. The manuscript does not report the distribution of reversal levels or how many staircases reached the floor, so the quantitative contrast between ICEV (1.93) and EV (3.48) may partly reflect a floor effect. The authors should report the raw reversal distributions and, ideally, use a finer step near threshold to verify that the ICEV advantage is not an artifact of censoring at the 2 km/h step.","section":"Procedure and Data Analysis (p.3), Table 2"},{"comment":"At 100 km/h, the ICEV audio stimulus was 63.7/63.6 dB(A) while the EV stimulus was 56.2/55.8 dB(A), a difference of about 7.5 dB(A). Overall loudness was not equalized across conditions. The Discussion addresses loudness only through the ratio of psychoacoustic metrics relative to each condition's own reference speed, which cancels absolute level differences. As a result, the observed JND difference between ICEV and EV could be driven by the substantially higher absolute sound level of the ICEV stimulus rather than by the presence of engine-order tones. Because the design recommendation (add speed-varying tonal components rather than increase loudness) depends on this attribution, the authors need to either match the overall level across conditions, include a level-matched control condition (e.g., EV sound with an added synthetic speed-varying tone, or ICEV sound with the tonal component removed), or otherwise provide evidence that absolute level does not account for the JND difference.","section":"Stimuli and Conditions (p.2), Table 1"},{"comment":"The central interpretation, that the ICEV advantage at highway speed is 'directly explained' by the tone-to-noise ratio, is an observational inference rather than a result of an experimental manipulation: the ICEV and EV stimuli differ in many acoustic attributes simultaneously, including overall level, spectral shape, roughness, and tonality. The frequency-shift consistency check also contains an apparent error: the text lists the second-order frequencies as 146.1 Hz and 145.0 Hz for engine speeds at 100 and 103 km/h, respectively, but the given RPM values (4350 and 4383) imply the opposite ordering (145.0 Hz at 100 km/h, 146.1 Hz at 103 km/h). In addition, the tonality analysis in Figure 6 identifies a tone at 'approximately 140 Hz,' which does not match the 145-146 Hz value used in the calculation. The sign and magnitude of the frequency-slope argument should be corrected and presented consistently. More importantly, the claim that tonality is the primary cue would be considerably strengthened by an additional condition that manipulates tonality independently of other features (e.g., adding a synthetic speed-varying tone to the EV sound), which the current design lacks.","section":"Discussion (pp.5-6), tonality analysis and frequency-shift argument"}],"minor_comments":[{"comment":"The text 'Descriptive statistics are summarised in Table 1' appears just before Table 2 is presented; the table containing the JND means is labeled Table 1 in the text but Table 2 in the table itself. Please renumber consistently.","section":"Results (p.4), Table numbering"},{"comment":"The reported degrees of freedom F(2,28) and F(2,50) do not directly correspond to the stated sample sizes (30 participants, with 20 remaining after exclusions at 40 km/h). If these are Greenhouse-Geisser-corrected degrees of freedom, please state the uncorrected values and the epsilon estimates so the reader can verify the analysis.","section":"Results (p.4), ANOVA degrees of freedom"},{"comment":"The paper excludes all 40 km/h runs from the 10 participants who had one non-convergent staircase, leaving n=20, but the sample size used in each ANOVA and in the descriptive statistics is not explicitly stated. Please clarify the effective sample size per condition and whether the two-way repeated-measures ANOVA was run only on participants with complete data.","section":"Data Analysis (p.3), participant exclusion at 40 km/h"},{"comment":"The procedure describes a '2-down/1-up' rule combined with PEST step-size adjustments, but PEST traditionally uses a different decision rule (e.g., a likelihood-ratio based stopping rule). The combination should be described more precisely, including how the 2-down/1-up convergence point (approximately 70.7% correct) relates to the definition of JND used here, since readers may expect a 75% threshold.","section":"Procedure (p.3), staircase rule terminology"},{"comment":"The computed engine-order frequencies at 100 and 103 km/h appear to be swapped relative to the stated RPM values; this should be corrected to avoid confusing readers about whether the frequency increases or decreases with speed.","section":"Discussion (pp.5-6), frequency shift sign"},{"comment":"There are a few typos, including 'at least 1 years of driving experience' (should be '1 year') and 'with respect to the event to the speed difference' (likely 'with respect to the speed difference'). Please proofread the manuscript.","section":"Throughout, minor typographical issues"}],"recommendation":"major_revision","confidential_remarks":"The paper is a reasonable experimental contribution, and the open-code/open-data practice is commendable. The main concern is interpretational overreach relative to the design: the tonality claim is post-hoc and confounded by absolute level and staircase resolution. The authors should be encouraged to address the major comments head-on; if they can provide raw staircase data showing that the 1.93 km/h value is not a floor artifact and either add a level-matched condition or acknowledge the confound, the paper may become publishable. I do not see evidence of circularity or citation manipulation; the self-citations for the NVH framework are tooling references and are appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one thing you should know: the paper’s headline result — ICEV gives a 1.93 km/h speed JND at 100 km/h versus 3.48 for EV — cannot be produced by the staircase procedure as described. The method sets a 2 km/h minimum step at that speed; every reversal level should be an even integer offset, so the mean of the last six reversals cannot be 1.93. That leaves two possibilities: the actual step was finer than stated, or many ICEV staircases sat at the 2 km/h floor. Either way, the quantitative ICEV advantage is unresolved before you even touch the loudness confound.\n\nWhat is genuinely new: this is the first 2AFC staircase measurement of speed-discrimination thresholds under EV versus ICEV interior noise, with silence as a control. That is a worthwhile gap to fill, and the design is mostly sensible — fixed visual scene, counterbalancing, 30 participants, and a standard adaptive procedure. The paper also does a decent job reviewing the speed-bias literature and connecting it to JND.\n\nThe soft spots beyond the staircase issue: the ICEV stimulus was about 7 dB(A) louder than the EV at 100 km/h, and the paper does not equalize overall level. The tonality analysis compares loudness ratios relative to each condition’s own reference, but that does not rule out absolute loudness as a contributor. The claim that engine-order tonality “directly explains” the divergence is therefore over-strong for the current data. Also, the 40 km/h ICEV and EV JNDs (1.24 and 1.19) are right at the 1 km/h minimum step, so those are also near floor, though the pattern there (silence worse) is more robust.\n\nWho it’s for: NVH engineers and anyone designing synthetic sound for EVs. If the raw data check out, it gives a concrete target for speed-feedback cues. But the manuscript in its current form needs a major revision. The authors should report the actual reversal distributions, clarify the step-size rule, and either re-run with level equalization or at least treat level as a covariate. I would not cite the 1.93 number until that is settled.\n\nRecommendation: send it to peer review. The question is relevant, the method is standard, and the flaw is detectable and fixable — that is exactly what referees are for. A desk reject would waste a potentially useful datapoint.","headline":"The paper's promising JND design is undercut by an internal staircase inconsistency that puts the ICEV advantage at 100 km/h in doubt before the loudness confound even enters.","tokens_in":10056,"tokens_out":3463,"would_cite":false,"duration_ms":28971,"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":"This paper claims that at highway speed, the combustion-engine cabin sound lets drivers detect speed changes almost twice as finely as electric-vehicle sound (mean JNDs 1.93 vs 3.48 km/h), and that the advantage comes from an audible…","keywords":["speed perception","just-noticeable difference","electric vehicle interior noise","engine-order tonality","tone-to-noise ratio","driving simulator","psychoacoustics","adaptive staircase"],"falsifier":"Re-run the 100 km/h two-alternative forced-choice task with ICEV and EV stimuli matched in overall A-weighted level, and separately with the approximately 140 Hz engine-order tone filtered out of the ICEV sound while preserving loudness; the tonality explanation fails if the JND gap between powertrain conditions disappears or if removing the tone leaves the gap unchanged.","tokens_in":9097,"feed_emoji":"🚗","tokens_out":6148,"duration_ms":51872,"temperature":0.7,"pith_summary":"The paper tries to establish that a car's interior sound is not just a comfort feature but part of the sensory machinery drivers use to feel their speed. Measuring just-noticeable speed differences in a driving simulator, it finds that at 100 km/h drivers detect speed changes down to about 1.9 km/h with combustion-engine sound, about 3.5 km/h with electric-vehicle sound, and about 5.2 km/h in silence. The authors argue that the combustion-engine advantage comes from a discrete engine-order tone whose frequency rises with speed, providing a trackable auditory cue, while the electric-vehicle interior sound is too broadband and steady to support that fine discrimination. If true, this matters for EV design: simply making electric cars louder may not restore speed sensitivity; adding a speed-linked tonal component might.","feed_headline":"At 100 km/h, engine sound beats EV noise for speed sense","feed_subtitle":"Speed-change detection: 1.9 km/h with engine sound, 3.5 with EV, 5.2 in silence.","key_machinery":"The central object is the engine-order tonal component: a discrete spectral peak in the interior sound whose frequency is an integer multiple of engine rotational frequency. The paper identifies a second-order engine harmonic near 140 Hz at 100 km/h with a tone-to-noise ratio of about 6 dB, and uses classical frequency-discrimination limits to show that the $0.37\\,\\text{Hz}$ per (km/h) shift of this peak is just enough to support the measured $1.93$ km/h JND. This tone is what carries the argument: it is the only analyzed acoustic feature that clearly differs between ICEV and EV at highway speed and that changes predictably with speed.","core_discovery":"The paper reports that at a 100 km/h reference speed, the just-noticeable speed difference averaged 1.93 km/h with ICEV interior sound, 3.48 km/h with EV interior sound, and 5.15 km/h in silence, with all pairwise differences significant; at 40 km/h, ICEV and EV did not differ (1.24 vs 1.19 km/h) while silence was worse (2.72 km/h). The paper argues that the highway-speed advantage of the ICEV is not explained by speed-dependent changes in loudness or sharpness, because those psychoacoustic metrics stay near the detectability band for both vehicle types. Instead, tone-to-noise-ratio analysis reveals a prominent engine-order tone near 140 Hz at 100 km/h, with a level around 6 dB above the noise, whose frequency shifts by roughly 0.37 Hz per (km/h). The frequency shift corresponding to the measured ICEV JND is about 0.49%, squarely within typical frequency-discrimination limits near 150 Hz. The conclusion is that trackable spectral cues, not overall loudness, explain why combustion-engine sound supports finer speed discrimination on the highway.","pith_inferences":["A concrete design target follows from the paper's numbers: near 100 km/h, a tonal cue with a frequency slope of about 0.37 Hz per (km/h) around 140 Hz should be audible, though the authors do not derive this as an explicit specification.","The frequency-mapping argument predicts that speed changes should be even more detectable during acceleration, where the tone sweeps continuously rather than in steps; this can be tested with time-varying speed profiles, which the present constant-speed design does not cover.","The 40 km/h null result suggests tonality is not the only useful cue; if confirmed, EV sound design could focus tonal additions on highway speeds and rely on broadband variation at city speeds."],"forward_implications":["At highway speed, the soundscape determines how finely a driver can sense speed changes: combustion-engine sound roughly halves the JND relative to silence and nearly halves it relative to EV sound.","At 40 km/h, any audible feedback helps, but powertrain type does not matter, suggesting that tonal cues become decisive only when speed and engine order combine to produce a salient tone.","Simply amplifying EV interior noise will not restore highway speed sensitivity, because the EV's limitation is not overall level but the lack of a trackable tonal cue.","Adding a speed-contingent tonal component to EV interior sound may support speed perception while preserving comfort, which the paper frames as the actionable design implication."],"supporting_citations":[{"why":"Supplies the classical evidence that attenuating vehicle noise biases speed estimation, motivating the JND question.","marker":"[1–3]"},{"why":"Supplies evidence that quieter cabins yield poorer speed maintenance and larger speed variability, indicating a perceptual cost.","marker":"[4,5]"},{"why":"Supports the premise that ICEV acoustic signatures carry speed- and load-dependent cues used to calibrate perceived speed.","marker":"[7,8]"},{"why":"Provides the visual-vestibular finding that self-motion cues can change perceived speed without proportionally changing discrimination thresholds, motivating threshold measurement.","marker":"[19,20]"},{"why":"Supplies the NVH auralization method used to generate the ICEV and EV interior-sound stimuli.","marker":"[24]"},{"why":"Supplies the PEST adaptive-staircase method used to estimate each JND.","marker":"[27]"},{"why":"Supplies the outlier-trimming rule applied to reversal data before computing JNDs.","marker":"[28]"}],"fun_headline_variants":["Engine sound sharpens speed sense at 100 km/h: 1.9 vs 3.5 km/h JND","Why engine hum beats EV silence for speed discrimination on highways","Frequency shift, not loudness, drives speed JND advantage of engine sound","EV quietness impairs speed perception: JND 3.5 km/h vs 1.9 for engine"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the ICEV's speed-perception advantage comes from the pitch-carrying engine tone rather than from the fact that the ICEV sound was about 7 dB(A) louder at 100 km/h; the experiment did not equalise overall loudness across the sound conditions.","fun_headline_variants_meta":{"raw":{"variants":["Engine sound sharpens speed sense at 100 km/h: 1.9 vs 3.5 km/h JND","Why engine hum beats EV silence for speed discrimination on highways","Frequency shift, not loudness, drives speed JND advantage of engine sound","EV quietness impairs speed perception: JND 3.5 km/h vs 1.9 for engine"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000534,"raw_usage":{"total_tokens":2642,"prompt_tokens":1092,"completion_tokens":1550,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":708,"completion_tokens_details":{"reasoning_tokens":1453}},"tokens_in":708,"tokens_out":1550,"duration_ms":9989,"temperature":1.0,"reasoning_tokens":1453,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:31:31.461247+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the 100 km/h two-alternative forced-choice task with ICEV and EV stimuli matched in overall A-weighted level, and separately with the approximately 140 Hz engine-order tone filtered out of the ICEV sound while preserving loudness; the tonality explanation fails if the JND gap between powertrain conditions disappears or if removing the tone leaves the gap unchanged.","supporting_citations":[{"cited_title":"Velocity perception in a moving observer,","cited_arxiv_id":null,"evidence_quote":"Supplies the NVH auralization method used to generate the ICEV and EV interior-sound stimuli."},{"cited_title":"Influence of the size of the field of view on motion perception,","cited_arxiv_id":null,"evidence_quote":"Supplies the PEST adaptive-staircase method used to estimate each JND."}],"review_version":2}