Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Acoustic Non-Stationarity Objective Assessment with Hard Label Criteria for Supervised Learning Models

Guilherme Zucatelli, Ricardo Barioni, Gabriela Dantas

TL;DR

Acoustic non-stationarity assessment faces heavy computational demands from traditional INS measures. The authors introduce Hard Label Criteria (HLC), which yields a single global binary non-stationarity label per signal by region-wise INS evaluation and adaptive thresholds, enabling supervised models to estimate non-stationarity efficiently. They design NANSA, a transformer-based network with an ANS Encoder and Pattern Extractor trained with binary cross-entropy on HLC labels, plus a lightweight NANSALW variant. Across AudioSet, DCASE, and FSD50K, NANSA and NANSALW consistently surpass baselines in accuracy and reduce processing time by orders of magnitude relative to INS, enabling real-time, on-device non-stationarity assessment. This work provides a scalable framework for objective non-stationarity estimation with practical impact on speech, audio scene analysis, and related tasks.

Abstract

Objective non-stationarity measures are resource intensive and impose critical limitations for real-time processing solutions. In this paper, a novel Hard Label Criteria (HLC) algorithm is proposed to generate a global non-stationarity label for acoustic signals, enabling supervised learning strategies to be trained as stationarity estimators. The HLC is first evaluated on state-of-the-art general-purpose acoustic models, demonstrating that these models capture stationarity information. Furthermore, the first-of-its-kind HLC-based Network for Acoustic Non-Stationarity Assessment (NANSA) is proposed. NANSA models outperform competing approaches, achieving up to 99% classification accuracy, while solving the computational infeasibility of traditional objective measures.

Acoustic Non-Stationarity Objective Assessment with Hard Label Criteria for Supervised Learning Models

TL;DR

Acoustic non-stationarity assessment faces heavy computational demands from traditional INS measures. The authors introduce Hard Label Criteria (HLC), which yields a single global binary non-stationarity label per signal by region-wise INS evaluation and adaptive thresholds, enabling supervised models to estimate non-stationarity efficiently. They design NANSA, a transformer-based network with an ANS Encoder and Pattern Extractor trained with binary cross-entropy on HLC labels, plus a lightweight NANSALW variant. Across AudioSet, DCASE, and FSD50K, NANSA and NANSALW consistently surpass baselines in accuracy and reduce processing time by orders of magnitude relative to INS, enabling real-time, on-device non-stationarity assessment. This work provides a scalable framework for objective non-stationarity estimation with practical impact on speech, audio scene analysis, and related tasks.

Abstract

Objective non-stationarity measures are resource intensive and impose critical limitations for real-time processing solutions. In this paper, a novel Hard Label Criteria (HLC) algorithm is proposed to generate a global non-stationarity label for acoustic signals, enabling supervised learning strategies to be trained as stationarity estimators. The HLC is first evaluated on state-of-the-art general-purpose acoustic models, demonstrating that these models capture stationarity information. Furthermore, the first-of-its-kind HLC-based Network for Acoustic Non-Stationarity Assessment (NANSA) is proposed. NANSA models outperform competing approaches, achieving up to 99% classification accuracy, while solving the computational infeasibility of traditional objective measures.

Paper Structure

This paper contains 11 sections, 5 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Review of the INS Framework
  4. The Hard Label Criteria (HLC)
  5. NANSA Architecture and Training Criterion
  6. Experiments
  7. Datasets and Baseline Models
  8. Implementation Details
  9. Metrics and Statistical Analysis
  10. Results and Discussion
  11. Conclusion

Figures (3)

  • Figure 1: Sample spectrogram signals and corresponding INS values extracted from AudioSet eval dataset: Noisy Speech (a), Wooden Knock (b) and Blowing Wind (c).
  • Figure 2: The NANSA model diagram.
  • Figure 3: ROC curves and Area Under Curve (AUC) for acoustic non-stationarity assessment.