Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Robust Transcription: Exploring Noise Injection Strategies for Training Data Augmentation

Yonghyun Kim, Alexander Lerch

TL;DR

This study investigates the impact of white noise at various Signal-to-Noise Ratio levels on state-of-the-art APT models and evaluates the performance of the Onsets and Frames model when trained on noise-augmented data.

Abstract

Recent advancements in Automatic Piano Transcription (APT) have significantly improved system performance, but the impact of noisy environments on the system performance remains largely unexplored. This study investigates the impact of white noise at various Signal-to-Noise Ratio (SNR) levels on state-of-the-art APT models and evaluates the performance of the Onsets and Frames model when trained on noise-augmented data. We hope this research provides valuable insights as preliminary work toward developing transcription models that maintain consistent performance across a range of acoustic conditions.

Towards Robust Transcription: Exploring Noise Injection Strategies for Training Data Augmentation

TL;DR

This study investigates the impact of white noise at various Signal-to-Noise Ratio levels on state-of-the-art APT models and evaluates the performance of the Onsets and Frames model when trained on noise-augmented data.

Abstract

Recent advancements in Automatic Piano Transcription (APT) have significantly improved system performance, but the impact of noisy environments on the system performance remains largely unexplored. This study investigates the impact of white noise at various Signal-to-Noise Ratio (SNR) levels on state-of-the-art APT models and evaluates the performance of the Onsets and Frames model when trained on noise-augmented data. We hope this research provides valuable insights as preliminary work toward developing transcription models that maintain consistent performance across a range of acoustic conditions.

Paper Structure

This paper contains 5 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Experiments
  3. Exp. 1: Impact of Noise on Pre-Trained Systems
  4. Exp. 2: Effect of Noise Injection during Training
  5. Conclusion

Figures (3)

  • Figure 1: Inference results of the Onsets and Frames model and the Kong et al. model, evaluated on the white noise-injected MAESTRO test split across varying SNR levels.
  • Figure 2: Inference results for the Onsets and Frames model across different CNR values, evaluated on the white noise-injected MAESTRO test split at varying SNR levels.
  • Figure :