Composing with acoustic niches: A laboratory soundscape evaluation of sound installation reactivity Authors: Valérian Fraisse, Nadine Schütz, Coralie Vincent, Marcelo M Wanderley, Catherine Guastavino, Nicolas Misdariis Published: 2026 ABSTRACT AbstractSound installations can shape the soundscape of public spaces. However, few studies have examined their impact on perceptual evaluations, especially in relation to the pre-existing sound environment. We report on an exploratory study investigating the effect of sound installations on soundscape evaluation, specifically manipulating compositional and acoustic reactivity in laboratory settings. This research is part of a research-creation collaboration around Nadine Schütz’s Niches Acoustiques—a new permanent installation for a public square in Paris. Participants familiar with the public space (N = 37) evaluated soundscapes combining High-Order Ambisonics field recordings with simulations of the installation with different forms of reactivity, along different time contexts and associated space uses—amplifying, attenuating, or transforming its content in relation to the pre-existing sound environment. Results show that time context strongly modulates soundscape impact of scenario-specific compositions changing according to daytimes and weekdays. There is greater potential for improving soundscape’s pleasantness and variety in quieter, less unpleasant settings, highlighting the difficulty of enhancing pleasantness in unpleasant, traffic-dominated soundscapes. Moreover, acoustic reactivity can modulate these effects in relation to context: results suggest that following spectrotemporal trends might be more effective in noisy contexts, while occupying spectrotemporal niches might suit calmer ones. Together, these findings reveal how installation reactivity can meaningfully shape public spaces’ soundscapes.Subject terms: Neuroscience, Psychology, Psychology Introduction Urban auditory environments are usually dominated by mechanical sounds such as traffic, air conditioners, and construction noise. Traditionally, cities have focused on minimizing the negative impacts of these sounds1 by controlling levels and mitigating sound sources2. But in the past decades, soundscape interventions3 and more specifically, public space sound installations4,5 have garnered increased attention to diversify and enrich the affective qualities of urban environments6. However, research on the addition of curated content to urban environments remains limited, and further work is needed to better understand how sound installations can influence soundscapes (see refs. 7,8 for a review). Advancing this knowledge could help bridge the persistent academia-practice gap in soundscape studies9, rethink the role of sound in city design10,11 and identify effective strategies for soundscape interventions12. In turn, such insights could support artists—who generally do not have the scientific resources required to assess audience feedback13—in designing site-specific sound installations14. A key issue sound artists face in their creative work is the articulation of new sonic elements within pre-existing soundscapes. This can be particularly challenging in urban settings, where sound environments include multiple sonic layers with ever-shifting spatial, spectral, and temporal dynamics15,16. The notion of acoustic niches, derived from bioacoustics and acoustic ecology, provides a useful conceptual framework to address these challenges. An acoustic niche describes how animal species differentiate their vocalization—both spectrally and temporally—to coexist within mature ecosystems17,18. When applied to urban soundscapes, a niche corresponds to “an occurrence of a sound emission at the moment that is the most favorable and that offers a particularly well adapted place for its expression”19. In the design of soundscape interventions, this concept can be approached in several ways. Semantically, sonic elements can be introduced when they are best aligned with the site’s context and activities. For instance, integrating specific sound events at time periods when they are most relevant (e.g., time of the day, days of the week, etc.). Acoustically, sounds can be inserted within spectrotemporal niches. For instance, added sounds may emerge in quieter moments or in spectral region less occupied by existing sources. Spatially, distinct sonic fields (zones in which added sounds are audible) can be created, leaving large portions of the forecourt ‘untouched’ and giving people choices to navigate within a diversified soundscape. These approaches are central to Nadine Schütz’s Niches Acoustiques, a permanent sound installation soon to be deployed in the forecourt of the new Judicial Court of Paris (see Figure S2 in the supplementary materials). Rooted in an artistic intent to compose with and within the existing soundscape, the installation accounts for spatial, acoustic, and semantic dimensions of its setting through reactivity. In contrast with interactivity, which involves a reciprocal action between several actors of the same system20, “reactivity” refers here to a system that respond to external conditions, such as the state of the surrounding environment, without reciprocity. Specifically, the sound installation considers two intertwined forms of reactivity: scenario-specific compositions considering temporal dynamics of sound environment and space use, and real-time reactivity to the spectral and temporal features of the sound environment. Together, these forms of reactivity aim to foster a seamless integration of the installation, shaping a balanced and context-aware sonic presence. The present study is part of a research creation-collaboration21 to inform the design of Niches Acoustiques and evaluate its impact on the public space soundscape. Earlier stages of the project involved field recordings22, the development of a soundscape simulation tool for laboratory evaluation of the installation23, and a first laboratory study with composition sketches7. Results provided insights into how different composition strategies might affect soundscape familiarity and variety—showing, for example, that abstract sounds24 tend to affect soundscape evaluations more than referential sounds. However, the previous study evaluated different composition sketches against the same ambient background soundscape, as a first step, without considering reactivity. This study builds upon previous research on Niches Acoustiques by exploring more elaborated composition sketches (available in the supplementary materials, Audio Files 4, 5, and 6) that account for real-time acoustic reactivity to the surrounding sound environment. In particular, we compare different ways of implementing the concept of acoustic niches. We assess two key compositional strategies and evaluate the associated soundscape impacts: scenario-specific compositions and real-time acoustic reactivity. First, we aim to examine the impact of the installation across a set of soundscape scenarios—reflecting different times of day and days of the week, and the associated patterns of space use. This approach allows for new insights into the role of time context in shaping soundscapes in presence of a curated sound installation. While temporal variations are known to strongly influence soundscapes25–27, prior research has typically evaluated sound installations or other soundscape interventions using a single representative period (e.g., see refs. 7,8,28–32). Here we will investigate three compositions curated for contrasting scenarios representing soundscape variations in terms of time of the day and time of the week, specifically: Rush hour, on a weekday morning; Afterwork, on a weekday evening; and Quiet time, on a weekend late morning. Second, we also aim to compare several strategies for operationalizing acoustic reactivity into the installation. While previous research reports on sound installations responding to surrounding sound environments (see ref. 33 for a review), the effects of acoustic reactivity on soundscape evaluations remain unexplored. This leads to our two main research goals: investigate how public space users evaluate soundscape in the presence of sound installations 1) across scenarios reflecting temporal variations; and 2) when different forms of acoustic reactivity are implemented. As this is the first soundscape study on acoustic reactivity, we seek to develop an approach that could be transferred to other soundscape interventions. Starting from a given set of acoustic measurements corresponding to each scenario’s spectral—third-octave band equivalent levels—and temporal—real-time sound levels—trends, we devised three strategies for implementing acoustic reactivity: Data-driven Tracked reactivity, Data-driven Inverse reactivity, and Artistic reactivity. The rationale for these strategies is to compare two general approaches that implement the same reactivity settings regardless of context and composition content (Data-driven) with one that involves careful, context-dependent curation by the artist (Artistic). In the Data-driven Tracked approach, composition sketches are filtered so that they follow the spectrotemporal trends of the surrounding sound environment. In the Data-driven Inverse approach, compositions are filtered so that they follow spectrotemporal trends inverse to the surrounding sound environment–in other term, they occupy acoustic niches. Both Data-driven approaches involve the same reactivity parameters regardless of time context and of the contents of the compositions. In contrast, Artistic reactivity is tailored by the artist according to the scenario and composition. This more complex implementation requires direct human operation in relation to time context: reactivity parameters are curated to each scenario by the artist and applied differently along the layers of the composition. As public space soundscapes evolve over time25,26, we hypothesize that the soundscape effects of the sound installation might differ across scenarios. Based on previous laboratory studies7,8, we expect the sound installation to particularly affect soundscape ratings of familiarity and variety. Regarding acoustic reactivity, we expect inverse forms of reactivity to be more oppositional and have stronger effects on soundscape, particularly variety—as sonic content would emerge more easily from the pre-existing acoustic environment34—when compared to tracked reactivity. Overall, the soundscape impact of acoustic reactivity is expected to be smaller than the effect of the sound installation itself. We also expect to capture soundscape effects that had not been previously observed, as there are currently no studies that examine the soundscape effects of acoustically reactive, scenario-dependent sound installations. Results This section presents results from exploratory and confirmatory data analyses35 of the participants’ soundscape ratings and sound source dominance scores across three scenarios (Rush hour, Afterwork and Quiet time) and five conditions (Baseline Scenario without compositions; and four composition conditions: No Reactivity; Data-driven Inverse; Data-driven Tracked; Artistic). See the Methods section for details of the experimental design, procedure and analysis, and see ref. 36 for an exhaustive description of the study’s methodological development. We first compare the different soundscape scenarios to characterize the pre-existing soundscape(s) and verify that they capture a sufficiently diverse range of soundscapes on-site (Comparing Baseline Scenarios section). Then, for each scenario (Rush hour scenario, Afterwork scenario, and Quiet time scenario sections), we compare five different conditions (Baseline Scenario; installation with: No Reactivity; Data-driven Tracked reactivity; Data-driven Inverse reactivity; Artistic reactivity), to investigate the soundscape impact of the sound installation and its acoustic reactivity in relation to the scenario. Comparing baseline scenarios Sound source dominance scores and soundscape ratings for all three Baseline Scenarios without compositions are presented in Fig. 1 and Fig. 2. A repeated-measure MANOVA on the three conditions shows a significant difference between scenarios both in terms of sound source dominance (Modified ANOVA-type Statistic [MATS] ≈ 125.4, p < 0.001) and soundscape ratings (MATS ≈ 64.6, p < 0.001). Follow-up repeated-measures ANOVAs reveal significant differences between scenarios for sound source dominance scores (Traffic and Passers-by) as well as for soundscape ratings (Pleasant, Soothing, Loudness, Character, and Varied—Supplementary Table S1). Comparing Baseline Scenarios without compositions: median and interquartile range for the sound source dominance scores. Comparing Baseline Scenarios without compositions: median and interquartile range for the soundscape ratings. Post-hoc Wilcoxon signed-rank tests—Table 1—indicate that the dominances of Traffic noises and especially noises from Passers-by are significantly different across Baseline Scenarios (see Fig. 1). As expected, Traffic noises are overall dominant, but significantly less dominant in Quiet time than in the other 2 scenarios. Sounds from Passers-by highly vary across scenarios: they are more dominant in Afterwork than in Quiet time, and more dominant in Quiet time than in Rush hour. Other sound sources are negligible, median values being close to zero for Nature, Music, and Other sounds. Regarding soundscape ratings (see Table 1 and Fig. 2), Quiet time is more pleasant, more soothing, and less loud than both Rush hour and Afterwork scenarios. For scales related with familiarity (Appropriate, Character, and Coherent), median scores between 60 and 80 indicate that all Baseline Scenarios are fairly representative of the studied space. However, Afterwork better reflects the character of the space than either Quiet time or Rush hour. Further, Afterwork’s soundscape is significantly more varied than the other Baseline Scenarios, yet we did not detect significant differences regarding the Emergent item. Table 1Statistical significance in the change of the soundscape ratings and sound source dominance scores between scenarios: Benjamini-Hochberg Wilcoxon signed-rank tests (*: p < 0.05; **: p < 0.01; ***: p < 0.001), r effect size estimate and associated 95% limits for the confidence interval on the effect size based on percentile interval bootstrap (N = 5000)Baseline ScenariosRush hour – Quiet timeRush hour – AfterworkAfterwork – Quiet timeGroupItemr [CI]r [CI]r [CI]Sound source dominanceTraffic−0.55*** [−0.76, −0.29]−0.18 [−0.48, 0.01]−0.45** [−0.69, −0.16]Passers−by0.46*** [0.17, 0.69]0.86*** [0.82, 0.87]−0.78*** [−0.86, −0.64]Soundscape ratingsPleasant0.36* [0.05, 0.63]−0.22 [−0.52, −0.01]0.50** [0.21, 0.75]Soothing0.42** [0.12, 0.67]−0.30 [−0.60, 0.03]0.61*** [0.36, 0.79]Loudness−0.64*** [−0.83, −0.40]−0.06 [−0.39, 0.00]−0.67*** [−0.84, −0.44]Character−0.16 [−0.45, −0.01]0.40* [0.10, 0.65]−0.57*** [−0.76, −0.32]Varied−0.03 [−0.40, −0.01]0.54*** [0.26, 0.77]−0.45** [−0.70, −0.15]Negative values for effect size estimates indicate that the reference condition (the condition on the left, e.g. Rush hour for the Rush hour – Quiet time column) had higher median scores for that item. Statistical significance in the change of the soundscape ratings and sound source dominance scores between scenarios: Benjamini-Hochberg Wilcoxon signed-rank tests (*: p < 0.05; **: p < 0.01; ***: p < 0.001), r effect size estimate and associated 95% limits for the confidence interval on the effect size based on percentile interval bootstrap (N = 5000) Negative values for effect size estimates indicate that the reference condition (the condition on the left, e.g. Rush hour for the Rush hour – Quiet time column) had higher median scores for that item. Overall, scenarios effectively represent a variety of soundscapes in the same space, with differences across scenarios in terms of both sound source dominance—Traffic and Passers-by—and soundscape’s pleasantness, familiarity, and variety. Rush hour scenario Soundscape evaluations for Rush hour’s five conditions (Baseline Scenario; No Reactivity; Data-driven Inverse; Data-driven Tracked; Artistic) are presented in Figs. 3 and 4. A repeated-measures MANOVA on the different conditions reveals significant differences across conditions for both sound source dominance scores (MATS ≈ 383.2, p < 0.001) and soundscape ratings (MATS ≈ 103.3, p < 0.001). Follow-up repeated-measures ANOVA indicates significant differences across conditions for all sound sources except Traffic. For soundscape ratings, another repeated-measures ANOVA indicates significant differences related to familiarity (Character, Appropriate, and Coherent) and varied (Varied and Emergent)—Supplementary Table S2. Rush hour scenario: median and interquartile range for the sound source dominance scores. Rush hour scenario: median and interquartile range for the soundscape ratings. Results of post-hoc signed-rank Wilcoxon tests for sound sources are shown in Table 2. When compared to the Baseline Scenario, we observe drastic increases in Nature, Music and Other sources dominance for all composition conditions, while all composition conditions except No Reactivity led to moderate decreases compared to the Baseline Scenario’s dominance of noises from Passers-by. When comparing reactivity conditions (Data-driven Tracked; Data-driven Inverse; Artistic) with the No Reactivity condition, we see that Nature sounds are significantly less dominant for the Data-driven Tracked condition, while Music sounds are significantly less dominant for the Data-driven Inverse condition. In turn, Music is less dominant in the Data-Driven Inverse than in the other conditions with reactivity, and Nature is more dominant in the Data-Driven Inverse than in the Data-Driven Tracked condition. Table 2Rush hour scenarioRush hourPassers-byNatureMusicOtherComparison groupr [CI]r [CI]r [CI]r [CI]Bas. Sce. – No rea.−0.32 [−0.59, −0.03]0.71*** [0.53, 0.84]0.87*** [0.87, 0.87]0.58** [0.32, 0.78]Bas. Sce. – Dd. Inv.−0.50* [−0.72, −0.23]0.71*** [0.53, 0.83]0.83*** [0.76, 0.87]0.55** [0.28, 0.76]Bas. Sce. – Dd. Tra.−0.54** [−0.73, −0.29]0.65*** [0.42, 0.80]0.87*** [0.86, 0.87]0.51** [0.24, 0.73]Bas. Sce. – Artistic−0.41* [−0.66, −0.11]0.72*** [0.55, 0.84]0.85*** [0.81, 0.87]0.62** [0.37, 0.79]No rea. – Dd. Inv.−0.25 [−0.55, −0.02]0.02 [0.01, 0.36]−0.67*** [−0.81, −0.46]−0.18 [−0.49, −0.01]No rea. – Dd. Tra.−0.23 [−0.53, −0.01]−0.61*** [−0.82, −0.35]0.16 [0.01, 0.47]−0.35 [−0.62, −0.05]No rea. – Artistic−0.10 [−0.42, −0.01]0.23 [0.01, 0.54]0.08 [0.01, 0.40]0.17 [0.01, 0.48]Dd. Inv. – Dd. Tra.−0.06 [−0.39, 0.00]−0.42* [−0.69, −0.13]0.76*** [0.62, 0.85]0.19 [0.01, 0.49]Dd. Inv. – Artistic0.10 [0.01, 0.41]0.25 [0.02, 0.55]0.65*** [0.41, 0.82]0.12 [0.01, 0.43]Dd. Tra. – Artistic0.16 [0.01, 0.47]0.26 [0.02, 0.55]−0.25 [−0.55, 0.02]0.30 [0.03, 0.59]Statistical significance in the change of the sound source dominance ratings between conditions: Benjamini−Hochberg Wilcoxon signed-rank tests (*: p < 0.05; **: p < 0.01; ***: p < 0.001), r effect size estimate and associated 95% limits for the confidence interval on the effect size based on percentile interval bootstrap (N = 5000). Negative values for effect size estimates indicate that the reference condition had higher median scores for that item (Bas. sce: Baseline Scenario; No rea.: No reactivity; Dd. Inv.: Data-driven Inverse; Dd. Tra.: Data-driven Tracked). Rush hour scenario Statistical significance in the change of the sound source dominance ratings between conditions: Benjamini−Hochberg Wilcoxon signed-rank tests (*: p < 0.05; **: p < 0.01; ***: p < 0.001), r effect size estimate and associated 95% limits for the confidence interval on the effect size based on percentile interval bootstrap (N = 5000). Negative values for effect size estimates indicate that the reference condition had higher median scores for that item (Bas. sce: Baseline Scenario; No rea.: No reactivity; Dd. Inv.: Data-driven Inverse; Dd. Tra.: Data-driven Tracked). Results of post-hoc signed-rank Wilcoxon tests for soundscape ratings are shown in Table 3. When compared to the Baseline Scenario, all composition conditions except the Data-driven Tracked lead to significant decreases in ratings related to familiarity (Character, Appropriate, Coherence), while all composition conditions also lead to increases in ratings related to varied (Varied, Emergence) with the exception of the Data-driven Inverse condition—which only affected Emergence despite a moderate effect size for Varied. When comparing conditions implementing reactivity to the No Reactivity condition, we see that the Data-driven Tracked less affected the soundscape’s character while the Artistic condition less affected the soundscape’s character. Otherwise, the Data-driven Inverse more strongly affected the soundscape’s character and appropriateness than the Data-driven Tracked and Artistic conditions. Table 3Rush hour scenarioRush hourCharacterAppropriateCoherentVariedEmergentComparison groupr [CI]r [CI]r [CI]r [CI]r [CI]Bas. Sce. – No rea.−0.57** [−0.79, −0.32]−0.48* [−0.71, −0.20]−0.52** [−0.74, −0.25]0.50** [0.22, 0.74]0.43* [0.14, 0.67]Bas. Sce. – Dd. Inv.−0.68*** [−0.82, −0.48]−0.55** [−0.76, −0.28]−0.54** [−0.76, −0.29]0.37 [0.08, 0.62]0.58** [0.33, 0.77]Bas. Sce. – Dd. Tra.−0.32 [−0.58, −0.04]−0.30 [−0.57, −0.03]−0.32 [−0.59, 0.03]0.56** [0.30, 0.78]0.50** [0.22, 0.74]Bas. Sce. – Artistic−0.55** [−0.76, −0.29]−0.51** [−0.72, −0.24]−0.49** [−0.72, −0.21]0.66*** [0.44, 0.81]0.54** [0.27, 0.76]No rea. – Dd. Inv.−0.07 [−0.39, 0.01]−0.24 [−0.53, −0.01]−0.03 [−0.38, −0.01]−0.19 [−0.49, −0.01]−0.12 [−0.45, 0.01]No rea. – Dd. Tra.0.42* [0.12, 0.67]0.31 [0.04, 0.59]0.37 [0.07, 0.63]0.06 [0.00, 0.39]0.01 [0.01, 0.37]No rea. – Artistic0.43* [0.13, 0.70]0.15 [0.01, 0.47]0.17 [0.01, 0.49]−0.01 [−0.37, −0.01]0.08 [0.01, 0.41]Dd. Inv. – Dd. Tra.0.52** [0.24, 0.74]0.46* [0.16, 0.70]0.30 [0.03, 0.58]0.26 [0.02, 0.57]0.03 [0.01, 0.37]Dd. Inv. – Artistic0.42* [0.11, 0.68]0.36 [0.07, 0.63]0.16 [0.01, 0.47]0.26 [0.02, 0.56]0.00 [0.00, 0.38]Dd. Tra. – Artistic−0.15 [−0.45, −0.01]−0.22 [−0.51, −0.01]−0.20 [−0.51, −0.01]−0.13 [−0.46, 0.01]−0.02 [−0.37, 0.00]Change of the soundscape ratings between conditions. See Table 2 and Data analysis section for explanations. Rush hour scenario Change of the soundscape ratings between conditions. See Table 2 and Data analysis section for explanations. Afterwork scenario Evaluations for Afterwork’s five conditions are presented in Figs. 5 and 6. Again, we detect significant differences between conditions on both sound source dominance scores (MATS ≈ 502.4, p < 0.001) and soundscape ratings (MATS ≈ 109.5, p < 0.001), Follow-up ANOVAs indicating significant differences across conditions for all sound sources, as well as for soundscape ratings related to familiarity (Character, Appropriate, and Coherent) and the Varied item but this time not the Emergent item—Supplementary Table S3. Afterwork scenario: median and interquartile range for the sound source dominance scores. Afterwork scenario: median and interquartile range for the soundscape ratings. For the sound source dominance scores, post-hoc Wilcoxon signed-rank tests (see Table 4) reveal that all composition conditions lead to large increases in perceived dominance of Nature and Music sounds, while we observe medium masking effects from Traffic noise for the Data-driven Inverse condition as well as for noises from Passers-by for the Data-driven Inverse and No Reactivity conditions. Compared with the three other versions of the composition, Nature sounds are significantly less dominant and Music sounds are significantly more dominant in the Artistic condition. Music is also more dominant in the No Reactivity condition than in both Data-driven Tracked and inverse conditions. Table 4Afterwork scenarioAfterworkTrafficPassers-byNatureMusicOtherComparison groupr [CI]r [CI]r [CI]r [CI]r [CI]Bas. Sce. – No rea.−0.35 [−0.61, −0.05]−0.41* [−0.65, −0.12]0.81*** [0.70, 0.87]0.85*** [−0.87, −0.79]0.33 [0.04, 0.60]Bas. Sce. – Dd. Inv.−0.58* [−0.80, −0.31]−0.49* [−0.71, −0.20]0.86*** [0.82, 0.87]0.85*** [−0.87, −0.79]0.42 [0.14, 0.66]Bas. Sce. – Dd. Tra.−0.36 [−0.61, −0.07]−0.21 [−0.50, −0.01]0.86*** [0.82, 0.87]0.85*** [−0.87, −0.80]0.35 [0.06, 0.60]Bas. Sce. – Artistic−0.07 [−0.40, −0.01]−0.24 [−0.54, −0.02]0.80*** [0.70, 0.86]0.87*** [−0.86, −0.87]0.36 [0.07, 0.63]No rea. – Dd. Inv.0.13 [0.01, 0.44]−0.05 [−0.40, −0.00]0.12 [0.01, 0.44]−0.43* [−0.69, −0.14]0.20 [0.01, 0.50]No rea. – Dd. Tra.0.01 [0.01, 0.38]0.19 [0.01, 0.48]−0.11 [−0.42, −0.01]−0.43* [−0.69, −0.14]0.08 [0.01, 0.40]No rea. – Artistic0.30 [0.04, 0.59]0.32 [0.04, 0.58]−0.53** [−0.74, −0.26]0.51** [0.23, 0.74]0.13 [0.01, 0.444]Dd. Inv. – Dd. Tra.−0.04 [−0.39, −0.01]0.32 [0.03, 0.58]−0.21 [−0.51, 0.01]0.03 [0.00, 0.37]−0.14 [−0.47, −0.01]Dd. Inv. – Artistic0.36 [0.07, 0.63]0.27 [0.02, 0.55]−0.66*** [−0.81, −0.44]0.69*** [0.47, 0.84]−0.32 [−0.60, −0.04]Dd. Tra. – Artistic0.38 [0.08, 0.65]0.08 [0.00, 0.42]−0.51** [−0.74, −0.24]0.67*** [0.44, 0.85]0.21 [0.01, 0.50]Change of the sound source dominance ratings between conditions. See Table 2 and Data analysis section for explanations. Afterwork scenario Change of the sound source dominance ratings between conditions. See Table 2 and Data analysis section for explanations. For soundscape ratings, post-hoc Wilcoxon signed-rank tests (see Table 5) indicate that all composition conditions led to significant alterations of scales related to familiarity (Character, Appropriate, and coherent) when compared to the Baseline Scenario, similar to the impact of the composition in the Rush hour scenario. Yet, effects on familiarity are slightly stronger in this scenario, especially for the character item, and only the Data-driven Inverse condition led to a moderate increase in the varied item. For this scenario, we did not detect any significant difference on soundscape ratings between the different forms of reactivity implemented. Table 5Afterwork scenarioAfterworkCharacterAppropriateCoherentVariedComparison groupr [CI]r [CI]r [CI]r [CI]Bas. Sce. – No rea.−0.53** [−0.73, −0.26]−0.49* [−0.71, −0.21]−0.38* [−0.63, −0.08]0.31 [0.03, 0.58]Bas. Sce. – Dd. Inv.−0.77*** [−0.87, −0.62]−0.48* [−0.70, −0.20]−0.51* [−0.72, −0.24]0.51* [0.25, 0.72]Bas. Sce. – Dd. Tra.−0.76*** [−0.85, −0.62]−0.53* [−0.73, −0.26]−0.49* [−0.71, −0.21]0.38 [0.08, 0.65]Bas. Sce. – Artistic−0.59*** [−0.78, −0.34]−0.44* [−0.68, −0.15]−0.47* [−0.69, −0.18]0.31 [0.03, 0.59]No rea. – Dd. Inv.−0.16 [−0.46, −0.01]−0.04 [−0.37, 0.00]−0.11 [−0.41, −0.01]0.16 [0.01, 0.47]No rea. – Dd. Tra.−0.15 [−0.46, −0.01]−0.06 [−0.38, 0.00]−0.12 [−0.44, 0.01]0.12 [0.01, 0.44]No rea. – Artistic0.10 [0.01, 0.42]0.20 [0.01, 0.51]−0.13 [−0.42, 0.01]−0.09 [−0.42, 0.01]Dd. Inv. – Dd. Tra.−0.03 [−0.37, 0.00]−0.06 [−0.39, 0.00]0.08 [0.01, 0.40]−0.19 [−0.50, 0.01]Dd. Inv. – Artistic0.34 [0.05, 0.62]0.11 [0.01, 0.43]0.11 [0.01, 0.42]−0.19 [−0.49, 0.01]Dd. Tra. – Artistic0.34 [0.05, 0.63]0.18 [0.01, 0.48]0.11 [0.01, 0.43]−0.06 [−0.38, −0.01]Change of the soundscape ratings between conditions. See Table 2 and Data analysis section for explanations. Afterwork scenario Change of the soundscape ratings between conditions. See Table 2 and Data analysis section for explanations. Quiet time scenario Evaluations for Quiet time’s five conditions are presented in Figs. 7 and 8. As with the two other scenarios, we detect significant differences between conditions on both sound source dominance scores (MATS ≈ 684.7, p < 0.001) and soundscape ratings (MATS ≈ 99.7, p < 0.001). Follow-up repeated-measures ANOVAs indicate significant differences across conditions for all sound sources and for soundscape ratings related to pleasantness (Pleasant, Soothing) and varied (Varied and Emergence), as well as to the Character item—Supplementary Table S4. Quiet time scenario: median and interquartile range for the sound source dominance scores. Quiet time scenario: median and interquartile range for the soundscape ratings. Regarding sound source dominances, post-hoc Wilcoxon signed-rank tests (see Table 6) reveal once again large increases in dominances of Nature and Music sounds for all conditions when compared to the Baseline Scenario. Compositions also mask noises from Passers-by except for the Data-driven Tracked condition, and they mask Traffic noises except for the Data-driven Tracked and No Reactivity conditions. When comparing across composition conditions, we see that Nature sounds are overall less dominant in the Data-driven Tracked condition and more dominant in the Data-driven Inverse condition, with no significant differences between the Artistic and the No Reactivity conditions. Otherwise, Music sounds are substantially more dominant in the Artistic condition than in any other version of the composition, while they are significantly less dominant in the Data-driven Tracked condition compared to Data-driven Inverse and No Reactivity conditions. We also detect some increase in Other sound sources for the Data-driven Inverse condition when compared to the Data-driven Tracked and the Artistic conditions. Table 6– Quiet time scenarioQuiet timeTrafficPassers-byNatureMusicOtherComparison groupr [CI]r [CI]r [CI]r [CI]r [CI]Bas. Sce. – No rea.0.28 [0.02, 0.56]0.61** [0.38, 0.78]0.87*** [0.87, 0.87]0.86*** [0.81, 0.87]0.16 [0.01, 0.46]Bas. Sce. – Dd. Inv.0.45* [0.15, 0.69]0.48** [0.20, 0.70]0.87*** [0.87, 0.87]0.86*** [0.81, 0.87]0.33 [0.05, 0.60]Bas. Sce. – Dd. Tra.0.23 [0.01, 0.52]0.10 [0.01, 0.41]0.82*** [0.71, 0.87]0.84*** [0.77, 0.87]0.09 [0.01, 0.40]Bas. Sce. – Artistic0.53* [0.27, 0.74]0.38* [0.09, 0.65]0.86*** [0.83, 0.87]0.87*** [0.87, 0.87]0.14 [0.01, 0.46]No rea. – Dd. Inv.0.22 [0.02, 0.51]0.12 [0.01, 0.43]0.50** [0.21, 0.75]0.22 [0.01, 0.51]0.36 [0.06, 0.63]No rea. – Dd. Tra.0.09 [0.00, 0.41]0.56** [0.31, 0.75]−0.84*** [−0.87, −0.77]−0.67*** [−0.85, −0.43]0.37 [0.07, 0.63]No rea. – Artistic0.31 [0.03, 0.59]0.41* [0.11, 0.65]0.18 [0.01, 0.49]0.73*** [0.55, 0.86]0.25 [0.01, 0.54]Dd. Inv. – Dd. Tra.0.20 [0.01, 0.49]0.48** [0.20, 0.70]−0.87*** [−0.87, −0.87]−0.58*** [−0.79, −0.33]−0.58** [−0.76, −0.33]Dd. Inv. – Artistic0.01 [0.00, 0.37]0.28 [0.02, 0.58]−0.61*** [−0.79, −0.38]0.82*** [0.72, 0.87]−0.44* [−0.67, −0.14]Dd. Tra. – Artistic0.32 [0.04, 0.58]0.32 [0.04, 0.59]0.79*** [0.64, 0.87]0.87*** [0.87, 0.87]0.27 [0.02, 0.56]Change of the sound source dominance ratings between conditions. See Table 2 and Data analysis section for explanations. – Quiet time scenario Change of the sound source dominance ratings between conditions. See Table 2 and Data analysis section for explanations. Effects of conditions on soundscape ratings are quite contrasting with the effects observed in the two other scenarios. When compared to the Baseline Scenario, post-hoc Wilcoxon signed-rank tests (Table 7) reveal that all composition conditions but the Data-driven Tracked led to significant increases in ratings for pleasant, with Data-driven Inverse and Artistic conditions also leading to significant increases in ratings for Soothing. Regarding familiarity, all conditions but Data-driven Tracked led to decreases in ratings for Character, but no significant differences were detected for the Appropriate or Coherent items. For soundscape variety, we detect increases in ratings in Varied and Emergent items for Data-driven Inverse and Artistic conditions. When compared to the No Reactivity condition, only the Data-driven Tracked less affected the Character and Varied items, while both Data-driven Inverse and Artistic conditions led to a more Pleasant, Soothing and Emergent soundscape with decreased ratings for Character when compared to the Data-driven Tracked condition. Table 7– Quiet time scenarioQuiet timePleasantSoothingCharacterVariedEmergentComparison groupr [CI]r [CI]r [CI]r [CI]r [CI]Bas. Sce. – No rea.0.52** [0.25, 0.73]0.28 [0.02, 0.56]−0.45* [−0.70, −0.16]0.38* [0.09, 0.64]0.29 [0.03, 0.58]Bas. Sce. – Dd. Inv.0.73*** [0.55, 0.84]0.52** [0.24, 0.73]−0.67*** [−0.83, −0.45]0.50* [0.22, 0.74]0.49** [0.20, 0.72]Bas. Sce. – Dd. Tra.0.24 [0.02, 0.52]−0.02 [−0.38, −0.00]−0.20 [−0.51, −0.01]−0.08 [−0.41, −0.01]0.08 [0.01, 0.40]Bas. Sce. – Artistic0.68*** [0.47, 0.83]0.55** [0.28, 0.76]−0.52** [−0.72, −0.24]0.41* [−0.66, −0.11]0.54** [0.27, 0.76]No rea. – Dd. Inv.0.27 [0.02, 0.57]0.17 [0.01, 0.47]−0.24 [−0.53, −0.02]0.07 [0.00, 0.39]−0.06 [−0.39, 0.00]No rea. – Dd. Tra.−0.30 [−0.59, −0.03]−0.35 [−0.62, −0.05]0.45* [0.17, 0.69]−0.43* [−0.68, −0.13]−0.30 [−0.57, 0.00]No rea. – Artistic0.15 [0.01, 0.47]0.17 [0.01, 0.49]−0.32 [−0.59, −0.04]−0.07 [−0.39, −0.01]−0.09 [−0.39, −0.01]Dd. Inv. – Dd. Tra.−0.45* [−0.71, −0.15]−0.47** [−0.70, −0.19]0.68*** [0.48, 0.82]−0.35 [−0.62, −0.06]−0.52** [−0.73, −0.26]Dd. Inv. – Artistic0.18 [0.01, 0.51]−0.06 [−0.39, 0.00]−0.00 [−0.37, 0.00]−0.11 [−0.43, −0.01]−0.16 [−0.47, −0.01]Dd. Tra. – Artistic0.44* [0.16, 0.70]0.49** [0.20, 0.72]−0.59*** [−0.77, −0.35]0.35 [0.05, 0.62]0.48** [0.18, 0.71]Change of the soundscape ratings between conditions. See Table 2 and Data analysis section for explanations. – Quiet time scenario Change of the soundscape ratings between conditions. See Table 2 and Data analysis section for explanations. Discussion This laboratory study explored how public space users evaluate their soundscape in the presence of the Niches Acoustiques sound installation across different scenarios representing time contexts and the associated patterns of use, and across different implementations of acoustic reactivity. The findings indicate that the soundscape effects of a scenario-dependent (i.e., changing according to daytimes and weekdays) sound installation are strongly tied to time context, and that acoustic reactivity can modulate these effects to a smaller extent than time context. Consistent with previous research, the sound installation consistently impacted soundscape familiarity (Character, Appropriate, Coherent) and variety (Varied, Emergence), with varying effects on pleasantness (Pleasant, Soothing)7,8,28,37,38. In line with previous laboratory studies7,8, this highlights the need to consider soundscape evaluation beyond the commonly-used two-dimensional pleasantness-eventfulness model. Specifically, the findings call for including scales relative to familiarity when investigating soundscape interventions, such as (but not limited to) the item for “appropriateness” in the Method A questionnaire in ISO 12913-239. As discussed in previous laboratory studies7,8, familiarity appears to play a key role in assessing soundscapes in the presence of unusual sounds, such as with the introduction of sound installations (see Axelsson and colleagues’ discussion on the matter40). However, an unprecedented finding is the strong difference in soundscape effects of the sound installation across scenarios. Results indicate that the sound installation was more likely to increase soundscape pleasantness and variety when evaluations for these dimensions were not already polarized in the Baseline soundscapes: pleasantness was increased only for the least unpleasant scenario—Quiet time—while variety was less affected in scenarios that had already high scores for this dimension—specifically Afterwork. On the one hand, this confirms that there might be stronger potential for increasing soundscape pleasantness and variety (and, by extension, eventfulness) when introducing sounds to quieter, less busy soundscapes28,38. This is to be nuanced, as a previous field study also showed that increasing pleasantness might not be achievable if the pre-existing soundscape is already perceived as highly pleasant37. On the other hand, this brings additional evidence for the difficulty of increasing pleasantness by introducing new sounds into an already unpleasant, traffic-dominated soundscape7,8. Furthermore, there was more potential for the installation to affect soundscape familiarity when the pre-existing soundscape had initial high ratings for this dimension—in the Rush hour and Afterwork scenarios. We believe that these findings might transfer to other contexts, but this needs to be further confirmed, as the respective influences of the scenarios and of the compositions (curated to the scenarios) on the observed effects cannot be disentangled in the present study. Results also highlight the ambivalence of the familiarity soundscape dimension: in Quiet time, the installation significantly modified soundscape’s character with no detectable effect on appropriateness or coherence items. This ambivalence, which corroborates qualitative findings from the previous study on Niches Acoustiques7, can have crucial implications on methodological and practical grounds. This provides support for assessing soundscape interventions’ impact on familiarity not only on the basis of the highly-connotated Appropriate item28,41, as suggested in the current ISO 12913-239, but rather complement it with the Character item (see Soundscape evaluation section). This can be critical in the design of soundscape interventions, since it is possible to affect a soundscape’s character without making it less appropriate (eventually demonstrating the feasibility of “rupturing” the monotony of urban soundscapes6). Regarding soundscape variety, detectable effects on Varied and Emergence are not always associated as well. Although this might reveal a limitation in statistical power (size effects sometimes indicating small to medium effects when pairwise tests do not reach significance), this provides further evidence that variety and emergence might not always be linked8. A soundscape intervention could therefore affect variety without influencing the presence of emerging sounds, and inversely. Further research is required to better understand the relationship between these items and in regards to eventfulness, which was not included in this study as it was assumed to also encompass the perceived presence of sources denoting human activities (e.g., voices, but also traffic or machines, see ref. 42). Regarding sound sources dominance, our findings reveal a small attentional masking effect of compositions on existing sound sources, namely a small decrease in perceived dominance of Traffic and Passers-by, which was more often observed on less dominant sound sources (Passers-by) than on dominant ones (Traffic). This is in line with previous soundscape studies in field37 or laboratory settings8, where sound installations were more likely to mask non-dominant sounds than dominant ones. Overall, these results demonstrate the importance of considering the time context (e.g. examining different periods such as different times of the day, or days of the week), when examining the impact of soundscape interventions: the space’s soundscape evolved substantially across time-based scenarios (in line with previous research25–27,37), along with the effects of the scenario-dependent sound installation. Regarding reactivity, we observed a significant impact of acoustic reactivity on sound source dominance scores and soundscape ratings across conditions and for each scenario, independently of perceived loudness. The effects of acoustic reactivity (i.e., when comparing the different versions of the composition) were of a smaller magnitude than the effects of the compositions themselves: we posit that acoustic reactivity can modulate the soundscape impact of the sound installation. Overall, we found that the effect of acoustic reactivity differs across scenarios, as with the effects of the sound installation itself. For soundscape ratings, compositions with Data-driven Inverse reactivity had a more pronounced impact on soundscape when compared to compositions with Data-driven Tracked reactivity: with inverse reactivity, we found higher alterations of scales related to familiarity in all three scenarios, as well as to pleasantness and variety for Quiet time. This confirms our prior intuition that an inverse approach—i.e., occupying acoustic niches—leads to a more oppositional installation34, with added sounds that tend to more strongly affect the pre-existing soundscape. When compared to the Inverse approach, the Tracked approach would appear to better fit in noisier situations, where a tight balance has to be struck between added sounds and the pre-existing noise7,8, especially given the small—although non-significant—deterioration of soundscape pleasantness with inverse reactivity for the Rush hour condition. These results suggest that a uniform approach to acoustic reactivity, independent of composition and temporal context, appears less effective. Instead, Artistic reactivity, adjusted to time context and compositions, proved more valuable, with less marked impacts on soundscape in Rush hour and Afterwork scenarios, and stronger soundscape enhancement in Quiet time. Otherwise, if the absence of (detectable) effects of reactivity on variety for Afterwork can be explained by the already highly varied pre-existing soundscape, for Rush hour it might be explained by the compositions themselves, with smaller differences in sound source dominance scores across conditions. In practical grounds, this study demonstrates the productivity of an approach to reactivity curated to representative time periods. For sound installations, sound artists are uniquely positioned to implement reactivity as expert listeners with an embodied knowledge of the site43. The principle of using a tracked approach in more noisy periods versus an inverse approach in quieter settings should transfer to other soundscape interventions, but further research is needed to validate this in a wider range of contexts. Finally, we found that reactivity always had stronger effects on sound source dominance scores than on soundscape ratings, with strong differences across reactivity conditions for Nature, Music and Other sounds. Amid current efforts to standardize approaches to design and assess soundscape interventions12, this contributes to the evidence that sound source dominance can capture unique effects that can be helpful for design purposes21. However, the scenario-specific interplay between the different layers of the compositions and reactivity makes it difficult to identify patterns that could be generalized beyond this study. The only general observation is that Nature sounds tend to be more dominant in inverse approaches when compared to the tracked approach, which might be related to the nature of the spectral filter applied to the conditions, reinforcing or diminishing the presence of the spectrally rich middleground layer (see Added sounds composition section). Furthermore, acoustic reactivity generally did not have a detectable effect on sound source masking. This exploratory study draws from multiple fields (acoustics, soundscape studies, sound installation art) and its methodological development was guided by an iterative, collaborative process between the research team and the sound artist36. Accordingly, rather than employing a clearly delineated factorial design (e.g., see refs. 44,45), we opted for a subset of conditions that balanced theoretical relevance with the artistic and practical goals of the project. Specifically, each composition was designed for a scenario reflecting specific time periods and was not relevant for others, such that effects of the compositions and of time context cannot be disentangled in the present study. We also combined spectral and temporal forms of reactivity as minimal differences were observed when they were implemented separately36. Altogether, the compositions tested can be seen as design prototypes rather than strictly controlled stimuli46. In addition, we tested multiple dependant variables (looking for potential directions that the effects of the intervention may take) and emphasized statistical power in our analyses47. This increases the risk that some results may not replicate in future confirmatory studies. Overall, our results can be considered provisional, awaiting confirmation on independent datasets35. Notwithstanding these limitations, the study allowed us to formulate high-level hypotheses paving the way for further research to test finer-grain hypotheses related to the effects we uncovered here (e.g., further investigations of the impacts of spectral and temporal reactivity in relation to the acoustic, psychoacoustic, or timbral features of the added sounds38,48,49). Pending field validation, we also believe that the underlying principles identified here could be transferred and tailored to other contexts: 1) a soundscape intervention might more easily impact soundscape pleasantness and variety if the pre-existing soundscape is not already polarized in those dimensions; 2) artistically curated reactivity appears to work best, otherwise inverse reactivity approaches might tend to better fit in less busy and noisy soundscapes and vice versa. Future research should investigate other key features of sound installation art, including prolonged exposure (which might yield insights into the installation’s impact on soundscape familiarity as people get used to it), socio-demographic factors, impact on people unfamiliar to the space, and other sensory stimulations such as visuals (e.g., see refs. 28,50), smells, or thermal comfort. The articulation of the sound installation in regard to users’ motion and behaviors is another important avenue for future research, especially in relation to the spatial niche concept explored in the artwork. This could be investigated through behavioral observations (e.g., see refs. 51,52) and other social architecture approaches (e.g., see ref. 53). Furthermore, ecological validity was only addressed in terms of representativity of the studied population—of participants familiar with the public space—but the experimental task, which involved listening to short excerpts with a high-level of listening attention, may not be representative of the cognitive process used in real-life conditions, which typically involves passive listening18. This was likely to magnify participants’ reception of the soundscape intervention, especially in laboratory settings26. The study design might also have led to potential mismatches with listening conditions in situ. For instance, interview feedback suggests that the scenarios might not have been representative of what all participants would do in the forecourt. The repeated-measures design might also have led to a form of repetitiveness with the identification of a clear narrative between conditions of the same scenario8. It should also be noted that the purpose of the experiment was not disclosed to participants to minimize potential bias stemming from prior opinions about sound installation art or socially desirable feedback13. A more reflexive approach could be considered to complement this perspective, for instance, involving the participants to reflect on the artistic project itself or even contribute to it, in an ethnographic approach (e.g., see refs. 54,55). This paper focused on soundscape scales ratings and sound source dominance scores, but qualitative data was also collected from all participants as part of follow-up semi-structured interviews. These interviews were mostly focused on a form of reactivated listening involving moments that were significant to participants during the experiment (see ref. 8). This data provides crucial information for interpretative guidance to make adequate design recommendations to the artists. These qualitative results, as well as installation design outcomes, will be reported in a forthcoming publication. To conclude, this study demonstrated that the influence of scenario-dependent sound installations—and, by extension, soundscape interventions—on soundscape pleasantness, familiarity, and variety is strongly tied to time context and the associated patterns of use. We found that improvements in pleasantness and variety are most easily achieved in soundscapes that are not already highly unpleasant or varied. Regarding familiarity, changing the soundscape’s Character appears to be more effective in highly familiar soundscapes, at the risk of decreasing appropriateness. We posit that these observations might hold for other soundscape interventions, but the compositions assessed here remained time-dependent, such that this would need further confirmation. The findings also highlight the value of an artistically curated approach to acoustic reactivity. At a more general level, we observed that reactivity strategies following the spectrotemporal dynamics of the pre-existing sound environment appeared to be more effective in less pleasant, traffic-dominated contexts, whereas inverse strategies, occupying spectrotemporal niches, seemed more suitable in quieter, less unpleasant soundscapes. Observed effects of reactivity on soundscape ratings remained subtle, and these tendencies need further validation in the field or using more controlled stimuli. While some of these findings may be transferable to other sound installations and soundscape interventions, public space sound art remains inherently site-specific, shaped by the unique spatial, acoustic, and social characteristics of each location. We therefore advocate for integrating similar research-creation methodologies into the design and evaluation of sound installations21, to enhance our understanding of how sound shapes public space experience. Methods Researchers and sound artist designed this study together and all aspects of the experimental apparatus are the result of joint decisions. The experiment uses a soundscape simulation tool implemented in Cycling ’74 Max 856, previously developed and validated through listening tests7,23. Experiment parameters, including scenarios’ design and reactivity settings were determined and refined through joint listening sessions between the two first authors and two pilot studies (N = 9) detailed in a separate paper36. All audio signals involved in the study have a 24-bit depth and a sample rate of 48 kHz. The study design, illustrated in Fig. 9, comprises four steps detailed in the following sections. Overall, the experiment assesses soundscapes across three scenarios that reflect temporal variations and associated patterns of space usage (Baseline Scenarios section), comparing them in the presence of compositions curated to the scenarios (Section Added sounds composition) with and without different forms of reactivity to conditions without added sounds, resulting in five conditions per scenario (Reactivity implementation and conditions section) for a total of fifteen conditions. Flowchart of the research design, modified from ref. 36. The sound environment (left) is simulated from HOA recordings. The installation’s monophonic compositions (right) are processed through a module simulating acoustic reactivity using laboratory measurements (center), then auralized with a 3D modeling of the space. Both the sound environment and added sounds are reproduced in the listening room via HOA streams for soundscape evaluation (bottom). Baseline scenarios Variations of the pre-existing soundscape at the forecourt are presented through three listening scenarios edited from fourth-order Higher-Order Ambisonics (HOA) recordings (see Fig. 9): Rush hour, Afterwork, and Quiet time. The process for creating the scenarios is detailed in a separate paper36 and is outlined here for the Reader’s convenience. Six scenarios were originally created and three of them were selected following pilot studies. Each scenario is linked to a 3 minutes 10 s HOA excerpt representative of a time of the day and a day of the week and associated with a narrative specific to the time represented (see Table 8 and Supplementary Audio Files 1, 2, and 3). Narratives were created based on fieldnotes and qualitative data from the previous study7: the forecourt is mainly used for transit purposes, either for commuting to or from work or for crossing the space, people occasionally sitting for short breaks (see Supplementary Fig. S1). Table 8Scenarios narratives and general descriptionNameNarrativeDescriptionRush hourIt’s a weekday morning, around rush hour. You’re on your way to workDominant traffic; few other sound sources; a few isolated human voices; quite loudAfterworkIt’s early evening on a weekday. You wait for a moment near the metro entrance, chatting with your colleagues or watching passers-by before heading home.Dominant human presence (footsteps, suitcases, voices); average traffic noise; moderately loud; eventfulQuiet timeIt’s late morning on a weekend. You take a stroll and stop on the forecourt for a break.Low traffic noise; few passers-by; some leisure activities (jogging, scooter); uneventfulNarratives are translated from French by the first author. Scenarios narratives and general description Narratives are translated from French by the first author. Recordings were collected during a measurement campaign in spring 2022, which included 10-min HOA recordings with an MH Acoustics Eigenmike 32 and sound level measurements with a B&K 2250 for calibration, placed at the approximate ear height for a seated person (1.3 m). Material used in this study were based on recordings selected from a larger database (see ref. 36), collected across three measurement sessions and four measurement points. The selection of recordings was based on their representativity of spatial (e.g., walking vs. sitting areas, see supplementary Fig. S1) and temporal (i.e., different times of the day, days of the week) variations in the forecourt soundscape. Scenarios were created based on the recordings through listening sessions including the first author and the sound artist. Audio recordings were first screened and then edited following a process fully described in ref. 36. We first identified relevant sections according to their general tone and dominant sound sources (see Table 8). Then, we systematically removed sonic elements that could reinforce the identification of repetitions between different conditions in a given scenario (e.g., intelligible speech). Finally, we created excerpts of 3 min 10 s for each scenario. This length, corresponding to the conditions’ duration, was determined as optimal for proper narrative evolution of the scenarios and the compositions7,8 while being adequate for urban soundscape simulation based on the ISO/TS 12913-2:201839 recommendations. Audio files (W channel) and spectrograms for each scenario are available in the supplementary materials—Audio Files 1, 2, 3 and Figures S3, S4, and S5. Scenarios were introduced to the participants with narratives related to space use (see Table 8). Narratives were based on most relevant activities reported by using space users’ feedback in our previous study (where participants described their experience using the space in semi-structured interviews7). The space was mostly described as a transit space for people working in the area (such as lawyers and magistrates at the Judicial Court) or living in the neighborhood. Finally, each scenario was auralized (see Added sounds auralization section) for a specific listening area (see Fig. 10), selected as best suited to the narratives in terms of space use while representing variations across scenarios. Map of the forecourt showing the future sound installation’s speakers, scenarios’ area and orientation, and their associated recording positions. Added sounds composition This study is based on a simulation of the soundscape design for the specific site. The projected sound installation will be spatialized on several speakers distributed across four street light poles throughout the forecourt (see Fig. 10 and supplementary Fig. S2). Three-minute compositions were curated by the sound artist, prototyping the future installation for each of the three scenarios (see supplementary materials, audio Files 4, 5, and 6). Earlier compositions emphasized salient sound events; however, results from the previous experiment7 showed that a more continuous middleground consisting of textural sound materials was essential to ensure perceptual coherence and for cohesion with the contrasting sonic layers of the pre-existing sonic environment—for example, the persistent hum of the ring road versus the variable engine noise from Avenue de la Porte de Clichy (Fig. 10). Consequently, the compositions follow a dual-layered middleground/foreground structure. The goal of the foreground layers is to create a new auditory foreground to capture attention, thereby pushing the monotonous sounds of traffic predominating the existing soundscape into the background. They feature sound events that are sought to be salient and emergent due to their distinct narrative and semantic content, derived from recordings in nearby public spaces and more or less edited and reinterpreted in abstract and tonal forms7. In contrast, middleground layers correspond to intermediate auditory layers, to be inserted between the composed foreground and the background traffic hum. The aim of the middleground layers is to avoid sonic disconnections between the installation and the pre-existing sound environment, which was sometimes observed in the previous study7. They are made of stable, large band sounds, composed from natural sound textures recorded in the area (see below). The goal for these middleground layers is to act as a perceptual ‘glue’, i.e., to reinforce cohesion with the pre-existing sound environment. Below are a few examples of sound events included in the compositions (orders are not representative of the sounds’ relative presence or dominance in the compositions): Rush hour:middleground: falling water dropsforeground: children’s voices, bike wheel, a saxophone, synth pad and high synth Rush hour:middleground: falling water dropsforeground: children’s voices, bike wheel, a saxophone, synth pad and high synth middleground: falling water drops middleground: falling water drops foreground: children’s voices, bike wheel, a saxophone, synth pad and high synth foreground: children’s voices, bike wheel, a saxophone, synth pad and high synth Afterwork:middleground: water streamforeground: children’s voices, sparrows, an instrument evoking the piano, the passage of a streetcar Afterwork:middleground: water streamforeground: children’s voices, sparrows, an instrument evoking the piano, the passage of a streetcar middleground: water stream middleground: water stream foreground: children’s voices, sparrows, an instrument evoking the piano, the passage of a streetcar foreground: children’s voices, sparrows, an instrument evoking the piano, the passage of a streetcar Quiet time:middleground: rustling foliageforeground: an instrument evoking the guitar, sparrows Quiet time:middleground: rustling foliageforeground: an instrument evoking the guitar, sparrows middleground: rustling foliage middleground: rustling foliage foreground: an instrument evoking the guitar, sparrows foreground: an instrument evoking the guitar, sparrows See supplementary materials for the spectrogram and audio files of the monophonic compositions—Audio Files 4, 5, 6 and Figures S6, S7, and S8. Added sounds auralization The monophonic compositions of the future sound installation are auralized after reactivity implementation (see below), to simulate the sound installation’s propagation within the surrounding space. The auralization uses EVERTims framework57. It combines image source beamforming with the aid of a 3D model of the space for early reflections and a Feedback Delay Network58 for late reverberation. Overall, the auralization unit was fine-tuned and validated in a preliminary listening test reported in ref. 23. Although the geometry was derived from scale drawings, we lacked key acoustic parameters (e.g., reverberation time, surface absorption). We therefore relied on perceptual cues to fine-tune the auralization framework. As reported in ref. 23, four expert listeners in spatial audio adjusted the model through analytical listening (comparing HOA reference excerpts with auralized signals). Then, the most adequate auralization model was selected and validated in a preliminary listening test (n = 12). Reactivity implementation and conditions In this study, we sought to operationalize the concept of acoustic niches by implementing spectrotemporal acoustic reactivity of the sound installation. We applied spectral filters (equalizers) and temporal filters (automation envelopes) to each composition to mimic three general approaches to acoustic reactivity to the baseline sound environment: Data-driven Tracked reactivity, Data-driven Inverse reactivity, and Artistic reactivity. Whereas both data-driven approaches rely on a single set of reactivity parameters across scenarios, Artistic reactivity parameters are curated for each scenario and between composition layers (middleground / foreground). In all three strategies, spectral and temporal filters are combined to maximize differences across conditions, as minimal differences were observed when applying spectral and temporal filter separately in a pilot study36. The general process for implementing acoustic reactivity (based on the scenarios) is illustrated in Fig. 9 and synthesized below (see ref. 36 for a detailed explanation of the methodological development). Overall, it involves four steps: Acoustic data representative of listening conditions is collected for each scenario. Scenarios are recorded in the calibrated listening room (see Ambisonics reproduction and mastering section) at the sweet spot and ear level (1.3 m) using a B&K 2250 sound level meter. Acoustic data representative of listening conditions is collected for each scenario. Scenarios are recorded in the calibrated listening room (see Ambisonics reproduction and mastering section) at the sweet spot and ear level (1.3 m) using a B&K 2250 sound level meter. Indicators relative to spectral and temporal acoustic energy are retrieved using Python packages Scipy and Numpy59. Indicators relative to spectral and temporal acoustic energy are retrieved using Python packages Scipy and Numpy59. Indicators are transferred to Reaper60 sessions using the Python package Reathon61. For each scenario, indicators are converted into automation envelopes (temporal reactivity) and 12-band equalizers (spectral reactivity). The associated Python code for reactivity implementation is available via this link: https://github.com/valerianF/analysis_codes/blob/main/npj_Acoustics_Composing_with_acoustic_niches/reactivity_implementation.py. Indicators are transferred to Reaper60 sessions using the Python package Reathon61. For each scenario, indicators are converted into automation envelopes (temporal reactivity) and 12-band equalizers (spectral reactivity). The associated Python code for reactivity implementation is available via this link: https://github.com/valerianF/analysis_codes/blob/main/npj_Acoustics_Composing_with_acoustic_niches/reactivity_implementation.py. The Reaper sessions are used by the artist to adjust spectral and temporal reactivities. Two unique sets of parameters across scenarios are chosen of the Data-driven Tracked and Data-driven Inverse approaches, while separate sets of parameters are selected for each scenario and across composition layers for the Artistic approach. The process for selecting and adjusting all reactivity parameters is detailed in ref. 36. The Reaper sessions are used by the artist to adjust spectral and temporal reactivities. Two unique sets of parameters across scenarios are chosen of the Data-driven Tracked and Data-driven Inverse approaches, while separate sets of parameters are selected for each scenario and across composition layers for the Artistic approach. The process for selecting and adjusting all reactivity parameters is detailed in ref. 36. Temporal reactivity is applied by applying automation envelopes to the added sounds in Reaper via Python. Automation envelopes are computed based on moving averages in the form of Decibel relative to full scale root mean square (RMSdBFS) values of the input signal s (corresponding to the scenario loops measurements with a sampling rate fs, here 48 kHz), with a specified window size δt, here 2 seconds36. This roughly corresponds to a normalized equivalent level of the pre-existing sound environment (LZeq,δt). For the tracked temporal reactivity, RMSdBFS values are multiplied by an intensity coefficient nt (see Eq. 1) while they are inverted and multiplied by nt to obtain the inverse temporal reactivity (Eq. 2). Values can also be time-shifted with a factor Δt. These temp values are sent into Reaper in the form of automation envelope points (in dB) with a refresh rate δt corresponding to the window size and a time-shift factor Δt of 2 s for Data-driven reactivities and 4 s for Artistic reactivity36. Then, points are interpolated in Reaper with a built-in piecewise cubic curve (corresponding to Reaper’s “Slow start/end” envelope point shape; see ref. 36 for detailed equations). Temporal intensity factors, shown in Table 10, were ultimately determined by the artist during listening sessions. Sound levels for the added sounds and the overall conditions are shown in Table 9. As expected, Inverse approaches exhibit less variability in the overall sound levels (LA10–LA90) than Tracked approaches since they “compensate” for variation in the pre-existing sound environment’s levels, while Tracked approaches “accentuate” these variations. Table 10Intensity factors for the spectral (nf) and temporal (nt) filters applied to the conditions with acoustic reactivityData-driven Tracked reactivityData-driven Inverse reactivityArtistic reactivityIntensity factorntnfntnfntnfComposition layerMG (Tracked)FG (Inverse)MG (Inverse)FG (Tracked)Rush hour33242263Afterwork33243133Quiet time33242262In contrast to the data-driven approaches, the Artistic reactivity conditions apply different intensity factors across scenarios and composition layers (MG: middleground; FG: foreground). Intensity factors for the spectral (nf) and temporal (nt) filters applied to the conditions with acoustic reactivity In contrast to the data-driven approaches, the Artistic reactivity conditions apply different intensity factors across scenarios and composition layers (MG: middleground; FG: foreground). Table 9Sound levels for each condition (Total) and for each version of the compositions without the Baseline Scenario (Added sounds only), measured under listening conditions in the calibrated listening room with a B&K 2250 at the sweet spot and seating hear level (1.3 m)ScenarioConditionTotal (Scenario + added sounds)Compositions onlyLAeq,3 minLA10 –LA90LoudnessLAeq,3 minLA10–LA90LoudnessRush hourBaseline Scenario60.14.213.8No Reactivity61.14.715.854.514.39.8Data-driven Inverse60.73.915.851.213.28.2Data-driven Tracked61.35.715.055.320.68.7Artistic61.15.015.054.222.28.7AfterworkBaseline Scenario60.26.713.2No Reactivity63.05.316.859.96.713.0Data-driven Inverse62.44.917.358.112.212.5Data-driven Tracked63.29.915.860.017.011.6Artistic62.46.915.258.316.410.8Quiet timeBaseline Scenario55.54.29.7No Reactivity56.44.711.549.511.66.9Data-driven Inverse56.54.211.850.013.56.3Data-driven Tracked55.74.59.942.815.03.8Artistic56.24.410.647.915.56.0Loudness corresponds to Zwicker’s loudness75. Sound levels for each condition (Total) and for each version of the compositions without the Baseline Scenario (Added sounds only), measured under listening conditions in the calibrated listening room with a B&K 2250 at the sweet spot and seating hear level (1.3 m) Loudness corresponds to Zwicker’s loudness75. Spectral reactivity consists of equalizers implemented in Reaper via Python that filter the added sounds according to levels measured for each scenario. Static filters were implemented under the assumption (based on previous measurements22) that the average spectrum of the pre-existing sound environment would only slightly fluctuate over short time periods. Equalizers were based on a modified version of Reaper’s 12-band parametric equalizer which uses biquad filters36,62, implemented with a Q factor of 1. The equalizer’s source code is available via this link: https://github.com/valerianF/analysis_codes/blob/main/npj_Acoustics_Composing_with_acoustic_niches/12-Band%20EQ%20Niches%20Acoustiques. To generate the spectral filters, A-weighted third-octave band equivalent levels (63 Hz–16000 Hz) were retrieved for each scenario. Corresponding octave-band levels (LAeq,band) were then computed every two third-octave bands between 80 Hz and 12500 Hz. These values were normalized by subtracting the logarithmic mean level (LAeq,mean) across all considered bands (63 Hz–16000 Hz) and divided by a spectral intensity factor (nf). The resulting values served as equalizer gains along their corresponding center frequencies (fc) for the tracked spectral filter (Eq. 3), while their negatives were applied for the inverse spectral filter (Eq. 4). As with temporal intensity factors, spectral intensity factors (see Table 10) were determined by the artist during listening sessions36. Overall and given the Baseline Scenarios’ spectral profiles, tracked filters have an overall bell shape (increased energy for mid frequencies; decreased energy for low and high frequencies) whereas inverse filters have and inverse bell shape (decreased energy for mid frequencies; increased energy for low and high frequencies), resulting in frequency levels as shown in Fig. 11, Fig. 12, and Fig. 13. Equalizer gains for each condition are available as supplementary Tables S5, S6, and S7. Third-octave equivalent levels (LAeq,3 min) of the conditions for the Rush hour scenario, measured under listening conditions in the calibrated listening room with a B&K 2250 at the sweet spot and seating hear level (1.3 m). Left: Total conditions (Baseline Scenario + added sounds). Right: Added sounds only. Third-octave equivalent levels (LAeq,3 min) of the conditions for the Afterwork scenario, measured under listening conditions in the calibrated listening room with a B&K 2250 at the sweet spot and seating hear level (1.3 m). Left: Total conditions (Baseline Scenario + compositions). Right: Added sounds only. Third-octave equivalent levels (LAeq,3 min) of the conditions for the Quiet time scenario, measured under listening conditions in the calibrated listening room with a B&K 2250 at the sweet spot and seating hear level (1.3 m). Left: Total conditions (Baseline Scenario + compositions). Right: Compositions only. Ambisonics reproduction and mastering The experiment was conducted at IRCAM’s studio 4, an acoustically treated listening room over a hemispherical dome of 24 Amadeus PMX 4 speakers placed at four height levels beginning at ear level (diameter ≈ 1.70 m). Encoding and decoding parameters were chosen during joint listening sessions as per the previous experiment (see ref. 7). Eigenmike 32 signals were encoded into a fourth-order HOA stream using spat∼63 with Tikhonov regularization64. The auralized added sounds and the sound environment—both as fourth-order HOA streams—were then summed and decoded in spat∼ using the energy-preserving method with max-rE weighting65. Baseline sound environment reproduction levels were calibrated using LAeq,5 min and LAeq,10 min field measurements, which were then compared with similar measurements in the listening room’s sweet sport at a height of 1.3 m, both using a B&K 2250 sound level meter. Similar to the previous laboratory experiment7, a −4 dB padding was then applied to better match the in situ perceived loudness66. For the added sounds, normalization and mastering were conducted to ensure consistent loudness across conditions within each scenario while mirroring the mastering process anticipated for on-site deployment. For each scenario, the following procedure was applied: Pre-normalization: Reference (No Reactivity) compositions were normalized in Reaper, adjusting the middleground and foreground layers separately to –30 LUFS, providing a baseline for the artist’s subsequent mastering. Pre-normalization: Reference (No Reactivity) compositions were normalized in Reaper, adjusting the middleground and foreground layers separately to –30 LUFS, providing a baseline for the artist’s subsequent mastering. Mastering: Three of the authors, including the artist, mastered the auralized reference (No Reactivity) compositions through listening sessions to replicate the future installation’s deployment process, again treating the middleground and foreground layers separately. Mastering: Three of the authors, including the artist, mastered the auralized reference (No Reactivity) compositions through listening sessions to replicate the future installation’s deployment process, again treating the middleground and foreground layers separately. Normalization of conditions with reactivity: Reactive conditions (Data-driven Tracked; Data-driven Inverse; Artistic) were normalized in Reaper relative to the corresponding mastered No Reactivity conditions, this time using overall LUFS-based normalization (with combined middleground and foreground), reflecting the intended automation of reactivity in the final installation. Normalization of conditions with reactivity: Reactive conditions (Data-driven Tracked; Data-driven Inverse; Artistic) were normalized in Reaper relative to the corresponding mastered No Reactivity conditions, this time using overall LUFS-based normalization (with combined middleground and foreground), reflecting the intended automation of reactivity in the final installation. Within each scenario, sound levels and Zwicker’s loudness varied to some extent across conditions (see Table 9). To assess whether these differences influenced loudness perception, we fitted a linear mixed model (using the R function lmer() from lmerTest package) with the Loudness soundscape scale as the dependent variable (see Questionnaire section), Zwicker’s loudness as a continuous predictor, scenario as a fixed factor, and participant as a random intercept (ICC ≈ 0.24). The test revealed no significant effect of Zwicker’s loudness on perceived loudness ratings (β ≈ 0.12, SE ≈ 0.59, t(515) ≈ 0.212, p ≈ 0.83), suggesting that these physical differences did not translate into detectable perceptual differences. Participants Thirty-seven participants were recruited for the experiment (age = 41.1 ± 13.9) based on their familiarity with the forecourt for ecological validity (representativeness of the studied population). All of them spoke French and self-reported normal hearing. These included both local residents (n = 18) and Judicial Court workers (n = 18), one participant being both a worker and a local resident. Participants were recruited through flyers distribution, recruitment ads posted in the neighborhood, CNRS’ RISC recruitment webform67, or word-of-mouth from co-workers. Participants received a 20€ compensation. Sample size was determined through an a priori power analysis for post-hoc Wilcoxon signed-rank tests (see Data analysis section) using G*Power version 3.1.9.768. Specifically, the power analysis indicated that a sample size of 35 would be sufficient to detect a phenomenon with a medium effect size of 0.5—based on data from the previous study7 and the pilot experiments—with an alpha level of 0.05 and a desired power of 0.8. This research has been approved by McGill Research Ethics Board (REB #22-01-033) and by Sorbonne University Research Ethics Committee (CER-2021-109). Informed consent was obtained from all participants. Procedure Participants were seated at the sweet spot of the hemispherical dome of 24 speakers and evaluated the different conditions using a Max interface on a 21.5-inch monitor using an external mouse. To contextualize listening, participants were first presented with a photograph, an aerial view, and a map of the forecourt in presence of a background sound environment (different from the Baseline Scenarios) for 60 s, while being asked to recall the space in their memory. Then, the following procedure applies for each of the three scenarios. First, the scenario’s narrative, Baseline excerpt and map were presented for 30 s. Then, participants were asked to listen to the five conditions associated to the scenario and, for each, evaluate both the perceived dominance of sound sources and the soundscapes using continuous scales—supplementary materials, Figure S9. Participants could optionally click on a button at any time during a trial to indicate the presence of a significant moment that struck them for a reason or another—supplementary material, Figure S9, top right. On each trial, the condition was presented for 3 min. A three-minute duration was determined to allow for sufficient narrative evolution of the compositions while being representative of the public space soundscape39. Participants listened for 10 seconds, then the questionnaire appeared, and they had 2 min 50 s to fill it out (we used momentary evaluations to reduce potential memory effects69). Participants were reminded to enter their ratings 30 s, 20 s, and 10 s before the end of a trial. To prevent abrupt silence and provide participants a brief rest, a 15-s low-level field recording served as a transition between conditions. We randomized the order of presentation of scenarios, conditions, and questionnaire items (except “Other sounds” that was always presented after the other sound source scales) to avoid order effects. Before the main experiment, participants provided written informed consent for data collection and publication. They were then given an instruction sheet—see supplementary Figures S10 and S11—and performed a supervised practice trial with a scenario different from the ones evaluated in the experiment, to familiarize themselves with the task. During that time, participants were free to ask questions, including for clarification about the meaning of the soundscape scales. Two optional breaks were automatically triggered between scenarios. At the end of the experiment, the experimenter conducted recorded semi-structured interviews, which will be presented in a forthcoming publication. Questionnaire The questionnaire comprises 13 continuous bipolar semantic scales, in French: five for sound source dominance from ISO/TS 12913-2:201839—Supplementary Table S8—and nine for soundscape evaluation validated in the previous study7 addressing soundscape pleasantness, variety, and familiarity—Supplementary Table S9. Pleasantness (Pleasant, Soothing, Sound Level) draws from PAQS39 and Welch et al.70. Familiarity (Character, Appropriate, Coherent) combines PAQS and PRSS items71 to measure the installation’s ability to reshape or reconfigure existing soundscapes6,7. Variety (Varied, Emergence) is conceptually linked to the eventfulness dimension in the PAQS but aims instead to capture perceived diversity, number, and motion of sources rather than human activity42. Data analysis Statistical analysis was performed using R v4.5.0 on RStudio 2025.05.0 with an alpha value of 0.05. Analysis code is available via this link: https://github.com/valerianF/analysis_codes/blob/main/npj_Acoustics_Composing_with_acoustic_niches/analysis_code.Rmd. Data were highly non-normal, with univariate and multivariate normality rejected for nearly all scales across scenarios, including residuals in univariate tests. Normality of pairwise group differences was also occasionally rejected. Homogeneity of variance was generally met, except for Character and Varied in the Afterwork condition, whereas homogeneity of covariance matrices was rejected for all subsets of scales (sound sources vs. soundscape evaluations) across scenarios. For these reasons, we chose to conduct semi-parametric and non-parametric analyses when pertinent. Analyses were conducted across four subsets of conditions. First, the three Baseline Scenarios (without added sounds) were compared to assess soundscape differences across scenarios. Then, the five conditions within each scenario were compared to examine the soundscape impact of composition and reactivity. Scales related to sound source dominance and soundscape evaluation were analyzed separately, as the former provides an analytical description of the sound environment, while the latter captures more holistic assessments. Both exploratory and confirmatory data analyses35 were performed when relevant. Specifically, in absence of preliminary hypothesis on the impact of sound installation reactivity on soundscape, all five conditions are compared with one another in a pairwise setting. Because of the exploratory nature of this research, we prioritize avoiding missing potentially important findings, i.e. maintaining some level of statistical power35. Thus, we rely on Benjamin-Hochberg’s p value correction method for multiple testing72. This method holds greater power than methods such as Bonferroni’s—which can be considered too conservative, especially with many tests—at the cost of increased risk of type I errors47. The following procedure was applied for each subset of conditions: A repeated-measure (RM) semi-parametric MANOVAs was conducted with either set of scales—sound source dominance or soundscape ratings—as dependent variables and the condition as independent variable, using the multRM function from the MANOVA.RM package version 0.5.373. Due to the covariance matrix being sometimes singular, we used the Modified ANOVA-type statistic (MATS) and wild bootstrap resampling method for p values with 10,000 iterations, as recommended by the package authors (see ref.74). A repeated-measure (RM) semi-parametric MANOVAs was conducted with either set of scales—sound source dominance or soundscape ratings—as dependent variables and the condition as independent variable, using the multRM function from the MANOVA.RM package version 0.5.373. Due to the covariance matrix being sometimes singular, we used the Modified ANOVA-type statistic (MATS) and wild bootstrap resampling method for p values with 10,000 iterations, as recommended by the package authors (see ref.74). Follow-up semi-parametric RM ANOVAs were conducted on each scale using the RM function from the same package, using the ANOVA-type statistic (ATS), again with wild bootstrap resampling method for p values with 10,000 iterations, using Benjamin-Hochberg p value correction. Each set of scales (sound sources' dominance scores and soundscape ratings) is considered separately for p value correction because they are assumed to cover different aspects of soundscape evaluation. Follow-up semi-parametric RM ANOVAs were conducted on each scale using the RM function from the same package, using the ANOVA-type statistic (ATS), again with wild bootstrap resampling method for p values with 10,000 iterations, using Benjamin-Hochberg p value correction. Each set of scales (sound sources' dominance scores and soundscape ratings) is considered separately for p value correction because they are assumed to cover different aspects of soundscape evaluation. For each scale that came out significant from the RM ANOVAs, we ran post-hoc pairwise (two-tailed) Wilcoxon signed-rank tests using the pairwise.wilcox.test function from the package stats (version 4.3.0) with Benjamin-Hochberg p value correction. The correction was applied to all pairwise comparisons, again considering each set of scales separately. We also report on r effect size estimates and the associated 95% confidence intervals based on percentile bootstrap with 5000 iterations, computed using wilcox_effsize function from the package rstatix version 0.7.2. For each scale that came out significant from the RM ANOVAs, we ran post-hoc pairwise (two-tailed) Wilcoxon signed-rank tests using the pairwise.wilcox.test function from the package stats (version 4.3.0) with Benjamin-Hochberg p value correction. The correction was applied to all pairwise comparisons, again considering each set of scales separately. We also report on r effect size estimates and the associated 95% confidence intervals based on percentile bootstrap with 5000 iterations, computed using wilcox_effsize function from the package rstatix version 0.7.2. Supplementary information Supplementary information Supplementary Audio 1 Supplementary Audio 2 Supplementary Audio 3 Supplementary Audio 4 Supplementary Audio 5 Supplementary Audio 6 Acknowledgements The authors are thankful to all the participants for giving their time and sharing their experience. The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: the study in this paper was funded by grants from the Social Sciences and Humanities Research Council of Canada [SSHRC, #890-2017-0065 to C.G., Sounds in the City], the Natural Sciences and Engineering Research Council of Canada [NSERC, Discovery Grant #2019-04551 to M.W.; Discovery Grant #2019-06121 and Acceleration Supplement #0035-2019 to C.G.]. Author contributions V.F.: Writing – review & editing, Writing – original draft, Visualization, Software, Project administration, Methodology, Formal analysis, Data curation, Conceptualization. N.S.: Writing – review & editing, Methodology, Conceptualization, Resources. C.V.: Writing – review & editing, Methodology, Investigation; M.W.: Supervision, Resources, Project administration, Conceptualization, Methodology; Funding acquisition; C.G.: Writing – review & editing, Validation, Supervision, Resources, Project administration, Conceptualization, Methodology, Funding acquisition; N.M.: Writing – review & editing, Supervision, Resources, Project administration, Conceptualization, Methodology, Funding acquisition. Data availability Due to the nature of consent obtained from participants, data cannot be made available outside of the research team, as per the ethics approval obtained from the McGill Research Ethics Board and the Sorbonne University Research Ethics Committee. For future data requests, please contact the McGill Ethics Officer at lynda.mcneil@mcgill.ca or the Sorbonne University Research Ethics Committee at CER@sorbonne-universite.fr. The codes that support the findings of this study are available via this link: https://github.com/valerianF/analysis_codes/tree/main/npj_Acoustics_Composing_with_acoustic_niches. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s44384-026-00066-5. References Associated Data Data Availability Statement Due to the nature of consent obtained from participants, data cannot be made available outside of the research team, as per the ethics approval obtained from the McGill Research Ethics Board and the Sorbonne University Research Ethics Committee. For future data requests, please contact the McGill Ethics Officer at lynda.mcneil@mcgill.ca or the Sorbonne University Research Ethics Committee at CER@sorbonne-universite.fr. The codes that support the findings of this study are available via this link: https://github.com/valerianF/analysis_codes/tree/main/npj_Acoustics_Composing_with_acoustic_niches.