Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CUHK-EE Systems for the vTAD Challenge at NCMMSC 2025

Aemon Yat Fei Chiu, Jingyu Li, Yusheng Tian, Guangyan Zhang, Tan Lee

TL;DR

A trade-off between model complexity and generalisation is highlighted, and the importance of architectural choices in fine-grained speaker modelling is underscored, pointing to future directions for improving robustness and fairness in timbre attribute detection.

Abstract

This paper presents the Voice Timbre Attribute Detection (vTAD) systems developed by the Digital Signal Processing & Speech Technology Laboratory (DSP&STL) of the Department of Electronic Engineering (EE) at The Chinese University of Hong Kong (CUHK) for the 20th National Conference on Human-Computer Speech Communication (NCMMSC 2025) vTAD Challenge. The proposed systems leverage WavLM-Large embeddings with attentive statistical pooling (ASTP) to extract robust speaker representations, followed by two variants of Diff-Net, i.e., Feed-Forward Neural Network (FFN) and Squeeze-and-Excitation-enhanced Residual FFN (SE-ResFFN), to compare timbre attribute intensities between utterance pairs. Experimental results demonstrate that the WavLM-Large+FFN system generalises better to unseen speakers, achieving 77.96% accuracy and 21.79% equal error rate (EER), while the WavLM-Large+SE-ResFFN model excels in the 'Seen' setting with 94.42% accuracy and 5.49% EER. These findings highlight a trade-off between model complexity and generalisation, and underscore the importance of architectural choices in fine-grained speaker modelling. Our analysis also reveals the impact of speaker identity, annotation subjectivity, and data imbalance on system performance, pointing to future directions for improving robustness and fairness in timbre attribute detection.

CUHK-EE Systems for the vTAD Challenge at NCMMSC 2025

TL;DR

A trade-off between model complexity and generalisation is highlighted, and the importance of architectural choices in fine-grained speaker modelling is underscored, pointing to future directions for improving robustness and fairness in timbre attribute detection.

Abstract

This paper presents the Voice Timbre Attribute Detection (vTAD) systems developed by the Digital Signal Processing & Speech Technology Laboratory (DSP&STL) of the Department of Electronic Engineering (EE) at The Chinese University of Hong Kong (CUHK) for the 20th National Conference on Human-Computer Speech Communication (NCMMSC 2025) vTAD Challenge. The proposed systems leverage WavLM-Large embeddings with attentive statistical pooling (ASTP) to extract robust speaker representations, followed by two variants of Diff-Net, i.e., Feed-Forward Neural Network (FFN) and Squeeze-and-Excitation-enhanced Residual FFN (SE-ResFFN), to compare timbre attribute intensities between utterance pairs. Experimental results demonstrate that the WavLM-Large+FFN system generalises better to unseen speakers, achieving 77.96% accuracy and 21.79% equal error rate (EER), while the WavLM-Large+SE-ResFFN model excels in the 'Seen' setting with 94.42% accuracy and 5.49% EER. These findings highlight a trade-off between model complexity and generalisation, and underscore the importance of architectural choices in fine-grained speaker modelling. Our analysis also reveals the impact of speaker identity, annotation subjectivity, and data imbalance on system performance, pointing to future directions for improving robustness and fairness in timbre attribute detection.

Paper Structure

This paper contains 11 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. System Description
  3. Audio Pre-Processing
  4. System Overview
  5. Speaker Encoder
  6. Diff-Net
  7. Experimental Settings
  8. Datasets
  9. Training Settings
  10. Results and Analysis
  11. Conclusion

Figures (3)

  • Figure 1: The overall design concept of the systems.
  • Figure 2: (a) The WavLM-Large module with the adoption of ASTP for voice feature extraction. (b) The original ASTP machanism.
  • Figure 3: (a) FFN-based Diff-Net. (b) SE-ResFFN-based Diff-Net. (c) An SE-ResNet block. (d) A squeeze-and-excitation block inside an SE-ResNet block.