Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Recognizing Ornaments in Vocal Indian Art Music with Active Annotation

Sumit Kumar, Parampreet Singh, Vipul Arora

TL;DR

This work tackles automatic recognition of vocal ornaments in Hindustani Art Music, a task hindered by scarce annotated data and fragile boundary handling. It introduces the Rāga Ornamentation Detection (ROD) dataset with six ornament types and a two-stage HITL annotation workflow, and proposes an Encoder-Decoder Temporal Convolutional Network (ED-TCN) that preserves ornament boundaries during chunking via a 'don’t-care' labeling scheme and periodic padding. The approach outperforms a CRNN baseline across diverse data splits and ragas, with strong cross-singer generalization and successful transfer to real concert recordings after limited fine-tuning. The results offer practical impact for music pedagogy, expressive singing synthesis, and robust ornament-aware MIR, while highlighting the need for broader datasets and domain adaptation to real-world performance conditions.

Abstract

Ornamentations, embellishments, or microtonal inflections are essential to melodic expression across many musical traditions, adding depth, nuance, and emotional impact to performances. Recognizing ornamentations in singing voices is key to MIR, with potential applications in music pedagogy, singer identification, genre classification, and controlled singing voice generation. However, the lack of annotated datasets and specialized modeling approaches remains a major obstacle for progress in this research area. In this work, we introduce Rāga Ornamentation Detection (ROD), a novel dataset comprising Indian classical music recordings curated by expert musicians. The dataset is annotated using a custom Human-in-the-Loop tool for six vocal ornaments marked as event-based labels. Using this dataset, we develop an ornamentation detection model based on deep time-series analysis, preserving ornament boundaries during the chunking of long audio recordings. We conduct experiments using different train-test configurations within the ROD dataset and also evaluate our approach on a separate, manually annotated dataset of Indian classical concert recordings. Our experimental results support the superior performance of our proposed approach over the baseline CRNN.

Recognizing Ornaments in Vocal Indian Art Music with Active Annotation

TL;DR

This work tackles automatic recognition of vocal ornaments in Hindustani Art Music, a task hindered by scarce annotated data and fragile boundary handling. It introduces the Rāga Ornamentation Detection (ROD) dataset with six ornament types and a two-stage HITL annotation workflow, and proposes an Encoder-Decoder Temporal Convolutional Network (ED-TCN) that preserves ornament boundaries during chunking via a 'don’t-care' labeling scheme and periodic padding. The approach outperforms a CRNN baseline across diverse data splits and ragas, with strong cross-singer generalization and successful transfer to real concert recordings after limited fine-tuning. The results offer practical impact for music pedagogy, expressive singing synthesis, and robust ornament-aware MIR, while highlighting the need for broader datasets and domain adaptation to real-world performance conditions.

Abstract

Ornamentations, embellishments, or microtonal inflections are essential to melodic expression across many musical traditions, adding depth, nuance, and emotional impact to performances. Recognizing ornamentations in singing voices is key to MIR, with potential applications in music pedagogy, singer identification, genre classification, and controlled singing voice generation. However, the lack of annotated datasets and specialized modeling approaches remains a major obstacle for progress in this research area. In this work, we introduce Rāga Ornamentation Detection (ROD), a novel dataset comprising Indian classical music recordings curated by expert musicians. The dataset is annotated using a custom Human-in-the-Loop tool for six vocal ornaments marked as event-based labels. Using this dataset, we develop an ornamentation detection model based on deep time-series analysis, preserving ornament boundaries during the chunking of long audio recordings. We conduct experiments using different train-test configurations within the ROD dataset and also evaluate our approach on a separate, manually annotated dataset of Indian classical concert recordings. Our experimental results support the superior performance of our proposed approach over the baseline CRNN.
Paper Structure (41 sections, 7 equations, 8 figures, 5 tables)

This paper contains 41 sections, 7 equations, 8 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Ornaments in Indian Art Music
  3. Kaṇ (grace notes)
  4. Mīṅd (gliding between notes)
  5. Murkī (subtle renditions of note clusters)
  6. Nyās svar (emphasized resting notes)
  7. Andolan (gentle oscillations)
  8. Gamak (heavy oscillations)
  9. Related Works
  10. Datasets for ornamentation detection task
  11. Analysis and Detection of Events
  12. Dataset Description
  13. Annotation Process
  14. Ornamentation-Specific Annotation Rules
  15. Inter-Annotator Agreement
  16. ...and 26 more sections

Figures (8)

  • Figure 1: Chromagram (top) and Pitch Contour (bottom) representations of ornaments in IAM (a): Kaṇ, (b): Mīnd, (c): Murkī, (d): Nyās svar, (e): Andolan, and (f): Gamak in ROD dataset. x-axis represents time (in sec.)
  • Figure 2: Class-wise ornament frequency for Singer 1 (left) and Singer 2 (right), with ornament classes on the x-axis and log-scaled counts on the y-axis. K, Me, An, Mu, H, and G represent Kaṇ, Mīnd, Andolan, Murki, Nyās Svar, and Gamak respectively.
  • Figure 3: An illustration of audio chunking using overlapping windows and corresponding label encoding. Label $C_n$ corresponds to class Andolan (within red lines) and $\dagger$ is the 'don't care' label.
  • Figure 4: An overview of the proposed methodology for the ornamentation detection task
  • Figure 5: Encoder-Decoder Temporal Convolutional Network
  • ...and 3 more figures