Soundscape Generation Using Web Audio Archives

The Research

The large and growing archives of audio content on the web have been transforming the sound design practice. In this context, sampling – a fundamental sound design tool – has shifted from mechanical recording to the realms of the copying and cutting on the computer. To effectively browse these large archives and retrieve content became a well-identified problem in Music Information Retrieval, namely through the adoption of audio content-based methodologies. Despite its robustness and effectiveness, current technological solutions rely mostly on (statistical) signal processing methods, whose terminology does attain a level of user-centered explanatory adequacy.

This dissertation advances a novel semantically-oriented strategy for browsing and retrieving audio content, in particular, environmental sounds, from large web audio archives. Ultimately, we aim to streamline the retrieval of user-defined queries to foster a fluid generation of soundscapes. In our work, querying web audio archives is done by affective dimensions that relate to emotional states (e.g., low arousal and low valence) and semantic audio source descriptions (e.g., rain). To this end, we map human annotations of affective dimensions to spectral audio-content descriptions extracted from the signal content. Retrieving new sounds from web archives is then made by specifying a query which combines a point in a 2-dimensional affective plane and semantic tags. A prototype application, MScaper, implements the method in the Ableton Live environment. An evaluation of our research assesses the perceptual soundness of the spectral audio-content descriptors in capturing affective dimensions and the usability of MScaper. The results show that spectral audio features significantly capture affective dimensions and that MScaper has been perceived by expert-users as having excellent usability.

The dissertation document will soon be published here:

This research led to the following paper:

  • Teixeira, P., Davies, M., Bernardes, G. (2019). The Online Musical Database in/as Performance. Hidden Archives, Hidden Practices: Debates About Music-Making. Aveiro, 2019.