Abstract A systematic review for soundscape modelling methods is presented. The methods for developing soundscape models are hereby questioned by investigating the following aspects: data acquisition methods, indicators used as predictors of descriptors in the models, descriptors targeted as output of the models, linear rather than non-linear model fitting, and overall performances. The inclusion criteria for the reviewed studies were: models dealing with soundscape dimensions aligned with the definitions provided in the ISO 12913 series; models based on soundscape data sampled at least at two different locations and using at least two variables as indicators. The Scopus database was queried. Biases on papers selection were considered and those related to the methods are discussed in the current study. Out of 256 results from Scopus, 22 studies were selected. Two studies were included from the references among the results. The data extraction from the 24 studies includes: data collection methods, input and output for the models, and model performance. Three main data collection methods were found. Several studies focus on the different combination of indicators among physical measurements, perceptual evaluations, temporal dynamics, demographic and psychological information, context information and visual amenity. The descriptors considered across the studies include: acoustic comfort, valence, arousal, calmness, chaoticness, sound quality, tranquillity, and vibrancy. The interpretation of the results is limited by the large variety of methods, and the large number of parameters in spite of a limited amount of studies obtained from the query. However, perceptual indicators, visual and contextual indicators, as well as time dynamic embedding, overall provide a better prediction of soundscape. Finally, although the compared performance between linear and non-linear methods does not show remarkable differences, non-linear methods might still represent a more suitable choice in models where complex structures of indicators are used.

中文翻译：

---

城市声景体验预测模型的系统回顾

摘要 对声景建模方法进行了系统回顾。开发音景模型的方法在此通过调查以下方面受到质疑：数据获取方法、用作模型中描述符预测器的指标、作为模型输出的描述符、线性而非非线性模型拟合以及整体性能。审查研究的纳入标准是： 处理音景维度的模型与 ISO 12913 系列中提供的定义一致；基于至少在两个不同位置采样并使用至少两个变量作为指标的音景数据的模型。查询了 Scopus 数据库。考虑了论文选择的偏见，并在当前的研究中讨论了与方法相关的偏见。在 Scopus 的 256 个结果中，选择了 22 项研究。结果中有两项研究来自参考文献。从 24 项研究中提取的数据包括：数据收集方法、模型的输入和输出以及模型性能。发现了三种主要的数据收集方法。一些研究侧重于物理测量、感知评估、时间动态、人口统计和心理信息、背景信息和视觉舒适度之间指标的不同组合。研究中考虑的描述词包括：声学舒适度、效价、唤醒度、平静度、混乱度、音质、宁静度和活力。尽管从查询中获得的研究数量有限，但结果的解释受到多种方法和大量参数的限制。然而，感知指标，视觉和上下文指标以及时间动态嵌入总体上提供了更好的音景预测。最后，虽然线性和非线性方法之间的比较性能没有表现出显着差异，但在使用指标结构复杂的模型中，非线性方法可能仍然是更合适的选择。