This short paper presents the technical foundations of Soundscapes of Warning (https://c2dh.uni.lu/soundscapes), an experimental open-source digital humanities platform for the comparative analysis, visualisation, and interactive exploration of public alarm signals. The system forms part of an ongoing project on the cultural and infrastructural history of warning sirens and develops methods suited to extremely short sonic artefacts. It has been extended to support cross-national comparison through a growing collection of sirens from different countries. Warning signals carry cultural meaning and historical density, yet their duration of only a few seconds poses challenges for both large-scale audio analysis and qualitative sound studies. The paper examines how the platform processes and transforms this type of sound material into interactive visual forms that allow users to select, listen to, inspect, and compare signals in relation to their geographic and historical contexts.

Central to the platform is the idea that this type of sound heritage and its semiotics require a different technical approach than longer recordings. These signals are significant precisely because they have been integrated into everyday life, where they have been naturalised and rarely treated as aesthetic or cultural objects (Schafer 1977; Bijsterveld 2008; Bull 2020). Each warning signal is approached as a complex sound object that can be re-mediated and placed in new contexts. By transforming these short sounds into interactive visual and spatial forms, and by situating them through metadata and geographic mapping, the system repositions alarms as objects of interpretation. This makes it possible to analyse how their meaning is shaped, encoded, and remembered in public listening (Cronqvist 2024).

The platform consists of three main components: audio preprocessing scripts in Node.js, a 3D visualization engine, and a browser-based geographic interface for comparison and interpretation. Each component is shaped by the short duration of warning signals, which requires specific choices in temporal resolution, spectral analysis, and the treatment of silence. The preprocessing pipeline extracts spectral features from short recordings gathered from national warning systems, archives, and field recordings. Given the brevity of the material, the system prioritizes comparability across samples through fixed FFT and window settings. Features such as amplitude, spectral centroid, and frequency distribution are stored as structured JSON, enabling direct rendering in the browser.

The visualization engine, implemented in React with TypeScript and a WebGL-based stack built on Three.js, transforms these features into three-dimensional forms derived from time, frequency, and amplitude. These forms function as interpretive models, allowing inspection of the internal structure of short sounds. The approach explores how visual representation can support the analysis of audio material, making variations in contour and intensity more perceptible. By avoiding interpolation, the system preserves the temporal brevity of the signals while adapting spectrographic conventions from phonetics, bioacoustics, and ecological listening (Joos 1948; Kroodsma 1989; Krause 2012). All visualizations are computed directly in the browser, enabling a lightweight, reproducible and easily deployable workflow. Notably, the development environment, built with Vite and React, allows researchers to run the website locally, execute the audio preprocessing pipeline, and instantly inspect the visual output, all without requiring a complex setup. Each signal is represented as a JSON object containing extracted features, geographic data, and other metadata, allowing researchers to inspect and refine its visual representation. The platform source code is hosted openly on GitHub (https://github.com/C2DH/soundscape), openly available to the public.

From a user experience perspective, the JSON object of each signal is mapped onto an interactive 3D globe. Selecting a geographically marked point loads the audio, displays its three-dimensional representation and its accompanying metadata. Metadata currently includes basic contextual information about codification, testing routines and device types, with planned expansions to integrate recording details and historical annotations. These layers help resituate short sounds within the infrastructures, practices and listening cultures that produce them (Cronqvist 2024; Malmstedt 2024).

The platform is an ongoing technical experiment exploring how digital humanities methods can support interpretive work with small sonic artefacts. By transforming a few seconds of sound into interactive visual forms and contextualising them geographically and historically, the system opens warning signals to close analysis, comparison and annotation. Rather than seeking definitive classifications, it creates conditions for exploratory listening and interpretive engagement. The work presented here represents the technical core of a larger project that integrates sound studies, historical research and experimental interface design to make the semiotics of public warning signals newly analyzable and discussable. This relatively simple setup supports meaningful exploration of sonic heritage without heavy infrastructure.