Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Oral Tradition-Encoded NanyinHGNN: Integrating Nanyin Music Preservation and Generation through a Pipa-Centric Dataset

Jianbing Xiahou, Weixi Zhai, Xu Cui

TL;DR

NanyinHGNN tackles preserving and innovating Nanyin’s heterophonic pipa-centered music by encoding core melodic material and oral ornamentations within a heterogeneous graph framework. It introduces a Pipa-Centric MIDI dataset and NanyinTok tokenization, plus a Graph Converter to convert symbolic sequences into graphs that respect modality, technique, and performance practices. The method unfolds in two stages: skeletal melody generation via an enhanced GATv2 and self-supervised ornamentation generation guided by oral tradition, with a rule-based ornamentation refinement stage. Experimental results show authentic four-instrument heterophonic ensembles, demonstrating that embedding domain knowledge into graph-based generation mitigates data scarcity in computational ethnomusicology.

Abstract

We propose NanyinHGNN, a heterogeneous graph network model for generating Nanyin instrumental music. As a UNESCO-recognized intangible cultural heritage, Nanyin follows a heterophonic tradition centered around the pipa, where core melodies are notated in traditional notation while ornamentations are passed down orally, presenting challenges for both preservation and contemporary innovation. To address this, we construct a Pipa-Centric MIDI dataset, develop NanyinTok as a specialized tokenization method, and convert symbolic sequences into graph structures using a Graph Converter to ensure that key musical features are preserved. Our key innovation reformulates ornamentation generation as the creation of ornamentation nodes within a heterogeneous graph. First, a graph neural network generates melodic outlines optimized for ornamentations. Then, a rule-guided system informed by Nanyin performance practices refines these outlines into complete ornamentations without requiring explicit ornamentation annotations during training. Experimental results demonstrate that our model successfully generates authentic heterophonic ensembles featuring four traditional instruments. These findings validate that integrating domain-specific knowledge into model architecture can effectively mitigate data scarcity challenges in computational ethnomusicology.

Oral Tradition-Encoded NanyinHGNN: Integrating Nanyin Music Preservation and Generation through a Pipa-Centric Dataset

TL;DR

NanyinHGNN tackles preserving and innovating Nanyin’s heterophonic pipa-centered music by encoding core melodic material and oral ornamentations within a heterogeneous graph framework. It introduces a Pipa-Centric MIDI dataset and NanyinTok tokenization, plus a Graph Converter to convert symbolic sequences into graphs that respect modality, technique, and performance practices. The method unfolds in two stages: skeletal melody generation via an enhanced GATv2 and self-supervised ornamentation generation guided by oral tradition, with a rule-based ornamentation refinement stage. Experimental results show authentic four-instrument heterophonic ensembles, demonstrating that embedding domain knowledge into graph-based generation mitigates data scarcity in computational ethnomusicology.

Abstract

We propose NanyinHGNN, a heterogeneous graph network model for generating Nanyin instrumental music. As a UNESCO-recognized intangible cultural heritage, Nanyin follows a heterophonic tradition centered around the pipa, where core melodies are notated in traditional notation while ornamentations are passed down orally, presenting challenges for both preservation and contemporary innovation. To address this, we construct a Pipa-Centric MIDI dataset, develop NanyinTok as a specialized tokenization method, and convert symbolic sequences into graph structures using a Graph Converter to ensure that key musical features are preserved. Our key innovation reformulates ornamentation generation as the creation of ornamentation nodes within a heterogeneous graph. First, a graph neural network generates melodic outlines optimized for ornamentations. Then, a rule-guided system informed by Nanyin performance practices refines these outlines into complete ornamentations without requiring explicit ornamentation annotations during training. Experimental results demonstrate that our model successfully generates authentic heterophonic ensembles featuring four traditional instruments. These findings validate that integrating domain-specific knowledge into model architecture can effectively mitigate data scarcity challenges in computational ethnomusicology.

Paper Structure

This paper contains 24 sections, 3 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Music Heritage Revival
  4. Skeleton-Guided Generation
  5. Expressive Performance and GNN Modeling
  6. Pipa-Centered Dataset
  7. From Score to Performance: A Practice Loop Grounded in Nanyin
  8. NanyinTok: A Token System for Modeling Playing Technique
  9. Graph Converter
  10. Heterogeneous Graph Structure
  11. Graph-Based Rules Injection
  12. Two-Stage Training Framework
  13. Stage 1: Skeletal Melody Generation with Enhanced GATv2
  14. Stage 2: Self-supervised Learning for Nianzhi Prediction and Rule-Guided Ornamentation
  15. Nianzhi Prediction
  16. ...and 9 more sections

Figures (2)

  • Figure 1: Red indicates liaopai rhythm markers, blue represents pipa playing techniques and pitch notation, large black characters show the song lyrics, and green denotes chen-ci (filler words).
  • Figure 2: Figure 1: The heterogeneous graph structure illustrates the relationships between musical notes (N), ornaments (O), and technical elements (green rectangles). Temporal connections are shown with dotted lines, decorative relationships with solid arrows, and trigger relationships with bold arrows. The nianzhi 3D feature vectors encode [is_nianzhi, speed_change, intensity_pattern] to capture the complexity of this traditional Chinese playing technique.