Dual-Glob Contrastive F0 Contour Framework
Figure 2: Overview of the proposed Dual-Glob framework. The model processes entire F 0 F_{0} contours via parallel clean ( x c x_{c} ) and augmented ( x a x_{a} ) views using a shared encoder. A composite supervised contrastive objective ( ℒ T o t a l \mathcal{L}_{Total} ) enforces structural consistency across both views to learn robust representations.
Paper context
Paper title: Deep Supervised Contrastive Learning of Pitch Contours for Robust Pitch Accent Classification in Seoul Korean Abstract: The intonational structure of Seoul Korean has been defined with discrete tonal categories within the Autosegmental-Metrical model of intonational phonology. However, it is challenging to map continuous $F_0$ contours to these invariant categories due to variable $F_0$ realizations in real-world speech. Our paper proposes Dual-Glob, a deep supervised contrastive learning framework to robustly classify fine-grained pitch accent patterns in Seoul Korean. Unlike conventional local predictive models, our approach captures holistic $F_0$ contour shapes by enforcing structural consistency between clean and augmented views in a shared latent space. To this aim, we introduce the first large-scale benchmark dataset, consisting of manually annotated 10,093 Accentual Phrases in Seoul Korean. Experimental results show that our Dual-Glob significantly outperforms strong baseline models with state-of-the-art accuracy (77.75%) and F1-score (51.54%). Therefore, our work supports AM-based intonational phonology using data-driven methodology, showing that deep contrastive learning effectively captures holistic structural features of continuous $F_0$ contours. Passages referencing this figure: re of Seoul Korean is hierarchically organized, with an Accentual Phrase (AP) as the basic unit, one or more of which are grouped into an Intonational Phrase (IP). According to Jun ( 1998 ) , APs with more than three syllables typically surface as LHLH or HHLH, depending on the phrase-initial segment: APs beginning with aspirated or tense consonants surface as HHLH, while others surface as LHLH ( Figure 1 ). Shorter APs show fourteen possible tonal patterns: LH, HH, LL, HL, LLH, LHH, HLH, LHL, HHL, HLL, LHLL, HHLL, LHLH, and HHLH (See Appendix A for the schematic contours of these patterns.). Intonational Phrase (IP) Accentual Phrase (AP) Word (W) σ \sigma T σ \sigma H (W) … \dots σ \sigma L (W) σ \sigma H (AP) % Figure 1: Intonational structure of Seoul Korean Jun ( 1998 ) . The AP-initia spirated or tense consonants surface as HHLH, while others surface as LHLH ( Figure 1 ). Shorter APs show fourteen possible tonal patterns: LH, HH, LL, HL, LLH, LHH, HLH, LHL, HHL, HLL, LHLL, HHLL, LHLH, and HHLH (See Appendix A for the schematic contours of these patterns.). Intonational Phrase (IP) Accentual Phrase (AP) Word (W) σ \sigma T σ \sigma H (W) … \dots σ \sigma L (W) σ \sigma H (AP) % Figure 1: Intonational structure of Seoul Korean Jun ( 1998 ) . The AP-initial tone (T) is realized as H for aspirated and tense consonants, otherwise L. The % symbol refers to a boundary tone (e.g., L% or H%) at the end of an IP. Despite this well-established theoretical characterization of intonation, there is a significant gap between phonological models Jun ( 1998 ) and real-world acoustic dat
