Reimagining Architecture: A Semiotic Study of Sound in Ai-Generated Spatial Design

This study examines how artificial intelligence (AI) interprets spoken architectural language by analysing vocal features—such as pitch, tone, and magnitude—and translating them into visual representations. Situated within the field of architectural semiotics, the research investigates how sound functions not merely as an acoustic phenomenon but as a symbolic agent in AI-mediated design. Five ambiguous architectural terms (vault, shell, column, plan, story) were recorded in two distinct sentence contexts and vocal styles (neutral vs. expressive). Using the Librosa Python library, pitch range and vocal magnitude were extracted as prosodic features. These metrics informed the construction of emotionally nuanced text prompts for MidJourney, a generative AI model, to produce architectural images reflecting vocal delivery. The results reveal consistent correlations between vocal variation and visual form: high pitch and strong vocal energy led to expressive, fluid, and emotionally charged spaces; lower pitch and stable magnitude generated grounded, monumental, or contemplative structures. These outcomes suggest that vocal expression can serve as a semiotic input in cross-modal AI workflows, where speech acts as both data and design material. By bridging sound and space, the study expands the semiotic framework of architectural representation and introduces voice as a generative modality in AI-assisted urban and spatial design. The findings support a multi-sensory design paradigm where not only what is said, but how it is said, shapes architectural meaning.