arXiv subjects

Hsing-Hang Chou

Publications and source records attributed to Hsing-Hang Chou.

2 recordsLinked to original sources

arXiv2025

ZSDEVC: Zero-Shot Diffusion-based Emotional Voice Conversion with Disentangled Mechanism

The human voice conveys not just words but also emotional states and individuality. Emotional voice conversion (EVC) modifies emotional expressions while preserving linguistic content and speaker identity, improving applications like human-machine interaction. While deep learning has advanced EVC models for specific target speakers on well-crafted emotional datasets, existing methods often face issues with emotion accuracy and speech distortion. In addition, the zero-shot scenario, in which emotion conversion is applied to unseen speakers, remains underexplored. This work introduces a novel diffusion framework with disentangled mechanisms and expressive guidance, trained on a large emotional speech dataset and evaluated on unseen speakers across in-domain and out-of-domain datasets. Experimental results show that our method produces expressive speech with high emotional accuracy, naturalness, and quality, showcasing its potential for broader EVC applications.

eess.AS↗

arXiv2025

Lessons Learnt: Revisit Key Training Strategies for Effective Speech Emotion Recognition in the Wild

In this study, we revisit key training strategies in machine learning often overlooked in favor of deeper architectures. Specifically, we explore balancing strategies, activation functions, and fine-tuning techniques to enhance speech emotion recognition (SER) in naturalistic conditions. Our findings show that simple modifications improve generalization with minimal architectural changes. Our multi-modal fusion model, integrating these optimizations, achieves a valence CCC of 0.6953, the best valence score in Task 2: Emotional Attribute Regression. Notably, fine-tuning RoBERTa and WavLM separately in a single-modality setting, followed by feature fusion without training the backbone extractor, yields the highest valence performance. Additionally, focal loss and activation functions significantly enhance performance without increasing complexity. These results suggest that refining core components, rather than deepening models, leads to more robust SER in-the-wild.

eess.AS↗