- This event has passed.
Paper Session 3b: Physiological and Physical Foundations of Creative Systems I
Session Chair: Tony de Ritis
Paper abstracts
Amir Abbas Orouji, Ayoub Banoushi and Gilberto Bernardes: “Vibrational Analysis of Traditional Persian Kamanche Sound Box: Experimental and Computational Investigation of Structural Modifications”
The kamanche, a bowed spike fiddle central to Persian classical music, features a spherical sound box covered with stretched animal skin and is played vertically on the performer’s lap. Despite acoustic similarities to the violin, comprehensive research on kamanche acoustics remains limited. This study investigates the acoustic contribution of the sound box to resonance characteristics and tonal quality of the closed-back kamanche, the most prevalent contemporary variant. The research combines COMSOL Multiphysics vibration simulation with experimental validation through impulse response frequency measurements. Investigated modifications include upper and lower hemisphere thickness variations and sound hole area reduction. Results demonstrate that upper hemisphere changes, while preserving internal air volume, substantially affect fundamental resonance patterns, corroborating traditional luthier observations. This study also suggests that the vibration modes 4,5, and especially 7 might be good candidates for maximum contribution to the overall amplifica-
tion of the string’s resonance and the overall sound of the instrument.
Nikolaus Knop: “Ponticello: An Interactive Conducting System for Mixed Music Performance”
Lucas Ong, Ruby Crocker and George Fazekas: “Emotion-Based Film Music Retrieval with Handcrafted and Deep Models”
Film music powerfully conveys emotion, yet computational methods for retrieving film tracks that match a target emotional state remain underexplored. This paper presents two approaches for emotion-based film music retrieval using Valence–Arousal (V–A) representations. The models are evaluated on the FME-24 dataset, which provides time-aligned participant-annotated V–A ratings for film music excerpts. The first approach applies k-Means to handcrafted audio features, while the second uses a VaDE model with contrastive learning to align audio and V–A embeddings. Results show that both methods capture emotion-related structure, with the deep model enabling more flexible, fine-grained selection.
