@incollection{RosenthalVersuemerSteffens2022, author = {Rosenthal, Fabian and Vers{\"u}mer, Siegbert and Steffens, Jochen}, title = {Audioinhaltsanalyse und Multilevelmodellierung zur Vorhersage der Bewertung von Indoor Soundscapes}, series = {Deutsche Jahrestagung f{\"u}r Akustik (DAGA)}, booktitle = {Deutsche Jahrestagung f{\"u}r Akustik (DAGA)}, publisher = {DAGA}, address = {Stuttgart}, year = {2022}, language = {de} } @misc{VersuemerSteffensRosenthal2023, author = {Vers{\"u}mer, Siegbert and Steffens, Jochen and Rosenthal, Fabian}, title = {Extensive crowdsourced dataset of in-situ evaluated binaural soundscapes of private dwellings containing subjective sound-related and situational ratings along with person factors to study time-varying influences on sound perception - research data}, edition = {V.01.1}, address = {Zenodo}, doi = {10.5281/zenodo.7193937}, year = {2023}, abstract = {The soundscape approach highlights the role of situational factors in sound evaluations; however, only a few studies have applied a multi-domain approach including sound-related, person-related, and time-varying situational variables. Therefore, we conducted a study based on the Experience Sampling Method to measure the relative contribution of a broad range of potentially relevant acoustic and non-auditory variables in predicting indoor soundscape evaluations. Here we present the comprehensive dataset for which 105 participants reported temporally (rather) stable trait variables such as noise sensitivity, trait affect, and quality of life. They rated 6.594 situations regarding the soundscape standard dimensions, perceived loudness, and the saliency of its sound components and evaluated situational variables such as state affect, perceived control, activity, and location. To complement these subject-centered data, we additionally crowdsourced object-centered data by having participants make binaural measurements of each indoor soundscape at their homes using a low-(self-)noise recorder. These recordings were used to compute (psycho-)acoustical indices such as the energetically averaged loudness level, the A-weighted energetically averaged equivalent continuous sound pressure level, and the A-weighted five-percent exceedance level. This complex hierarchical data can be used to investigate time-varying non-auditory influences on sound perception and to develop soundscape indicators based on the binaural recordings to predict soundscape evaluations.}, subject = {soundscape}, language = {en} } @masterthesis{Rosenthal2022, type = {Bachelor Thesis}, author = {Rosenthal, Fabian}, title = {Einsatz und Anpassung von Methoden der Audio-Feature-Extraktion am Beispiel von Indoor Soundscapes}, address = {D{\"u}sseldorf}, organization = {Hochschule D{\"u}sseldorf}, doi = {10.20385/opus4-4399}, url = {http://nbn-resolving.de/urn:nbn:de:hbz:due62-opus-43991}, school = {Hochschule D{\"u}sseldorf}, pages = {77}, year = {2022}, abstract = {Im h{\"a}uslichen Umfeld erleben Menschen abh{\"a}ngig von ihrer Wohnsituation unterschiedlichste Ger{\"a}uschumgebungen (Indoor Soundscapes), die einen hohen Einfluss auf das Wohlbefinden haben. Im Rahmen einer Feldstudie wurde daher die theoretisch und praktisch relevante Frage untersucht, welche Merkmale der wahrgenommenen Ger{\"a}uschumgebungen als besonders ereignisreich oder angenehm bewertet werden. Einhundertf{\"u}nf Teilnehmer*innen berichteten nach der Experience-Sampling-Methode zeitgesteuert {\"u}ber auftretende Ger{\"a}uschumgebungen des h{\"a}uslichen Alltags. Sie bewerteten deren subjektive Wirkung gem{\"a}ß Soundscape-Standard. Zudem fertigten sie in-situ Audioaufnahmen der Ger{\"a}uschszenarien an. Die 6594 Tonaufnahmen wurden einer Audioinhaltsanalyse unterzogen und im Zuge dessen wurden vier Featuresets verschiedener Berechnungsans{\"a}tze extrahiert. Mithilfe der Perzentilen LASSO-Regularisierung wurden lineare gemischte Modelle f{\"u}r Angenehmheit und Ereignisreichtum sowie drei Multilevel-Modelle f{\"u}r Angenehmheit aufgestellt. Die besten Modelle erkl{\"a}ren 9 \% der Varianz von Angenehmheit und 27 \% der Varianz von Ereignisreichtum durch feste Effekte selektierter Pr{\"a}diktoren aller getesteten Featuresets. Angenehmheit sinkt vor allem, wenn lautheitsbasierte Features hohe Werte zeigen. Ereignisreichtum ist am st{\"a}rksten abh{\"a}ngig von kurzen Spitzen des C-bewerteten Schalldruckpegels und wird im Vergleich zu Angenehmheit st{\"a}rker von Zeitschwankungen der Features bestimmt. Durch den Vergleich der Featuresets wird deutlich, dass die Modelleffekte bekannter psychoakustischer Gr{\"o}ßen durch Hinzuf{\"u}gen von MFCC-Features verbessert werden. Dar{\"u}ber hinaus wird anhand von Multilevel-Modellen gezeigt, dass die Angenehmheit des lautesten Viertels der Tonaufnahmen deutlich besser durch feste Effekte erkl{\"a}rbar ist als der leisere Rest des Datensatzes. Kongruent mit aktuellen Befragungsstudien kann durch Audiofeatures eine Kategorienabh{\"a}ngigkeit der Soundscape-Bewertungen belegt werden: Musik und Sprache werden als angenehmer bewertet, technische Ger{\"a}usche und Anlagenrauschen hingegen als unangenehmer. Hoher Ereignisreichtum h{\"a}ngt mit menschgemachten Ger{\"a}uschen (z. B. Poltern, Stuhlr{\"u}cken) zusammen. Diese Ergebnisse stellen eine wichtige Komponente in der umfassenden Beschreibung komplexer Ger{\"a}uschumgebungen dar und ihnen kommt in Zeiten der SARS-CoV-2-Pandemie, in denen Wohnraum vermehrt heterogen genutzt werden muss, eine erh{\"o}hte Bedeutung zu.}, language = {de} } @article{VersuemerBlaettermannRosenthaletal.2025, author = {Vers{\"u}mer, Siegbert and Bl{\"a}ttermann, Patrick and Rosenthal, Fabian and Weinzierl, Stefan}, title = {A comparison of methods for modeling soundscape dimensions based on different datasets}, series = {The Journal of the Acoustical Society of America}, volume = {157}, journal = {The Journal of the Acoustical Society of America}, number = {1}, publisher = {AIP Publishing}, issn = {1520-8524}, doi = {10.1121/10.0034849}, url = {http://nbn-resolving.de/urn:nbn:de:hbz:due62-opus-48849}, pages = {234 -- 255}, year = {2025}, abstract = {Soundscape studies vary considerably in study design, statistical methods, and model fit metrics used. Due to this confounding of data and methods, it is difficult to assess the suitability of statistical modelling techniques used in the literature. Therefore, five different methods and two performance metrics were applied to three existing soundscape datasets to model soundscape Pleasantness and Eventfulness based on seven acoustic and three sociodemographic predictors. Datasets differed in soundscape type (urban outdoor vs indoor), experimental setting (field- vs lab-based), size, and study design (site- vs person-centered). The fixed-effects and mixed-effects methods ranged from linear to nonlinear regression based on advanced machine learning approaches. Results showed that models performed better for Eventfulness than for Pleasantness in most cases, while performance as measured by the out-of-sample R2 was dependent on the total variance of the target, especially in both field studies with imbalanced targets and groups. Nonlinear methods consistently outperformed linear regression, with random forest and extreme gradient boosting performing particularly well, while the performance levels of all nonlinear methods remained comparable. Mixed-effects models provided a more generalized, albeit slightly smaller prediction performance when tested on unknown groups. Finally, this study motivates the use of cross-validation with special splitting for analyzing small imbalanced datasets.}, subject = {Akustische Eigenschaft}, language = {en} } @misc{RosenthalVersuemerBlaettermann2024, author = {Rosenthal, Fabian and Vers{\"u}mer, Siegbert and Bl{\"a}ttermann, Patrick}, title = {flexcv: Easy and flexible nested cross validation for tabular data in python}, edition = {Version v24.0-beta}, publisher = {Zenodo}, address = {Genf}, doi = {10.5281/zenodo.10160846}, year = {2024}, abstract = {Python package customizing nested cross validation for tabular data.}, subject = {Software}, language = {en} } @inproceedings{RosenthalBlaettermannVersuemer2024, author = {Rosenthal, Fabian and Bl{\"a}ttermann, Patrick and Vers{\"u}mer, Siegbert}, title = {flexcv: Python package for fitting, comparing, and logging multiple machine learning models using various cross-validation methods [Poster]}, series = {Fortschritte der Akustik - DAGA 2024, 18. - 21. M{\"a}rz 2024, Hannover}, booktitle = {Fortschritte der Akustik - DAGA 2024, 18. - 21. M{\"a}rz 2024, Hannover}, publisher = {Deutsche Gesellschaft f{\"u}r Akustik e.V.}, address = {Berlin}, year = {2024}, abstract = {The evaluation of listening experiments and studies on acoustics not only requires basic knowledge of statistics, but also poses challenges in the implementation of different methodologies. Especially for small sample sizes and peculiarities regarding hierarchical data structures, the need to individualize evaluation scripts arises. Therefore, we introduce flexcv, a powerful machine learning package for Python for evaluating various models on experimental tabular data, especially with small sample sizes. It supports random effects evaluation (including random slopes) for both linear and non-linear regressors, providing broad applicability to different experiments and research questions.flexcv quickly allows to perform nested cross-validation on a variety of models for comparison with each other. On the one hand, the implementation of a flexible interface simplifies the exchange of methods in the script, allowing researchers to change cross-validation methods without having to touch the actual cross-validation code. On the other hand, extensive online logging allows and simplifies the evaluation and experiment tracking along the process and different machines.}, language = {en} }