Data Augmentation Techniques for Convolutional Neural Network–Based Vocal Pitch Estimation in Carnatic Music

DOI

10.2991/978-94-6463-976-6_7How to use a DOI?

Keywords

vocal pitch estimation; Carnatic music; convolutional neural network; data augmentation; time stretching; pitch shifting; spectrogram enhancement; mel-frequency cepstral coefficients; constant-Q transform; dynamic thresholding; pitch contour estimation

Abstract

Vocal pitch estimation is an important part of analysis and preservation of Carnatic music where microtonal variations and ornamental movements define the melodic identity. However, the detection of accurate and stable pitch is still a challenge because of the diversity in vocal styles, recording conditions and the limited availability of annotated data. This paper proposes a conceptual framework for vocal pitch estimation based on convolution neural network architecture (CNN) with the help of data augmentation techniques to improve the model robustness and generalization. The proposed design makes use of time stretching, pitch change and spectrogram transformation to simulate various vocal expressions while retaining the tonal and temporal details of Carnatic music. Mel-frequency cepstral coefficients (MFCCs) are used as the basic input features for feature representation and constant-Q transform (CQT) is proposed as a complementary feature for the better harmonic information. An analytic dynamic-thresholding stabilization is proposed in order to stabilize the pitch contour and reduce the frequency jitter. The performance of the framework is shown in an illustrative way through analytical trends and expectations supported by literature, and it can be seen that the pitch-tracking accuracy, the contour transitions, and the noise tolerance are better than those of traditional rule-based approaches. This study gives a mathematically grounded and literature supported basis for the future implementation of CNN-based pitch estimation systems with the applications in music education, performance analysis and digital preservation of Indian classical traditions.

Copyright

© 2025 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Oral Roberts Nithiprakasam
AU  - S. Revathi
PY  - 2025
DA  - 2025/12/29
TI  - Data Augmentation Techniques for Convolutional Neural Network–Based Vocal Pitch Estimation in Carnatic Music
BT  - Proceedings of the International Conference on Intelligent Information Systems Design and Indian Knowledge System Applications (ICISDIKSA 2026)
PB  - Atlantis Press
SP  - 101
EP  - 119
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6463-976-6_7
DO  - 10.2991/978-94-6463-976-6_7
ID  - Nithiprakasam2025
ER  -