Akademik Veri Yönetim Sistemi
Electrica, cilt.26, 2026 (ESCI, Scopus, TRDizin)
This study introduces the expressive fluency analysis model (EFAM), a computational framework grounded in a geometric manifold interpretation of speech behavior. Each speaker's Z-normalized acoustic responses define a speaker-specific manifold in a 20-dimensional feature space; EFAM quantifies its structure through switching fluency (SF, trajectory length), domain coherence (within-domain compactness), and cross-domain adaptation (between-domain separation), unified in a composite Emotional Fluency Index (EFI). The framework was applied to 1530 speech recordings from 102 Azeri-Persian bilingual women responding to 15 clinical inventory items (GAD-7, PHQ-8) mapped onto five emotional domains. Within-speaker Z-normalization isolated relative expressive variation from individual vocal baselines. Hierarchical clustering revealed a somatic-cognitive domain bifurcation. Speaker-independent classification achieved 55.9% accuracy (chance = 2.0%), with MFCC-only models (60.3%) outperforming the full feature set and identifying vocal timbre as the primary acoustic channel for domain discrimination. Robustness was verified through permutation testing (SF canonical order at 1.1th percentile, P = .011), distance metric comparison (Euclidean vs. Mahalanobis: classification 58.5% vs. 52.5%; metric means nearly identical, ρ = 0.52–0.75), and bootstrap CIs. No significant age-related differences were found. The EFAM provides a reproducible geometric framework for modeling expressive organization in natural speech.