Ontologies of Sound in Neural Network Engineering

Contemporary research on AI and sound is predominantly led by engineers and computer scientists, often with non-music-related backgrounds. Their conceptualization of sound and music directly influences the algorithms they design, making it important to understand the ontological frameworks that inform their work. Using mixed-methods linguistic analysis on a corpus of arXiv articles, this study examines word occurrences and collocates in current neural network engineering literature, focusing on key notions like sound, audio, music, listening, nature, originality, and objectivity. The analysis reveals how machine learning practices tend to prioritize quantitative language, performance metrics, and un-situated approaches to auditory phenomena. In contrast to classical research on digital sound technologies, neural network engineering currently appears to be less grounded on phenomenological experience and more detached from artists and audiences.