Distribution-Preserving Output Restructuring for Ambient Visual Feedback in Speech Emotion Recognition: A Design-Oriented Comparative Study
IEEE Access · 2026 · 14 · 100017-100035
DOI: 10.1109/access.2026.3706867 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Kai Ma und 2 weiteren Autor:innen. Die Arbeit nennt 34 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Philosophy and Social Science Planning Fund Project of Zhejiang Province.
Abstract (Originalsprache)
Speech emotion recognition (SER) systems often produce probabilistic outputs, but interface-oriented applications commonly reduce these outputs to a single dominant label. This simplification hides ambiguity, category competition, and low-information states that may be meaningful for ambient feedback. This study investigates SER output organization as a presentation-layer design problem rather than as a classifier-optimization problem. Without modifying the underlying recognizer, we introduce a distribution-preserving output restructuring framework that organizes runtime SER probability outputs into three operational states: Sharp, Bimodal, and Flat. These states are mapped to ambient visual control parameters including color, intensity, diffusion radius, and propagation speed. Using a matched side-by-side prototype, we compared a deterministic dominant-category baseline with the proposed distribution-preserving strategy under shared input, hardware, color, and viewing conditions. A user study with 16 participants combined questionnaire ratings, open-ended responses, frame-level logs, and threshold sensitivity analysis. The distribution-preserving strategy did not show a uniform advantage across all measures, but it was most clearly favored on interpretive and expressive dimensions, including Interpretability, Perceived Match, and Mixed Emotion Expression. Comfort and Stability remained more variable. Log-based analyses supported the operational stability of the classification pipeline within the tested threshold range, while also clarifying that Flat states were strongly associated with low-energy or non-speaking intervals. The findings provide exploratory evidence that presentation-layer organization can shape how probabilistic SER output is perceived as ambient emotional feedback.
Förderung
- National Natural Science Foundation of China
- Philosophy and Social Science Planning Fund Project of Zhejiang Province
Zitieren
APA
Ma, K., Tang, Z., & Wang, Y. (2026). Distribution-Preserving Output Restructuring for Ambient Visual Feedback in Speech Emotion Recognition: A Design-Oriented Comparative Study. IEEE Access, 14, 100017-100035. https://doi.org/10.1109/access.2026.3706867
Vancouver
Ma K, Tang Z, Wang Y. Distribution-Preserving Output Restructuring for Ambient Visual Feedback in Speech Emotion Recognition: A Design-Oriented Comparative Study. IEEE Access. 2026;14:100017-100035. doi:10.1109/access.2026.3706867
BibTeX
@article{Ma2026,
author = {Ma, Kai and Tang, Zhichuan and Wang, Yu},
title = {Distribution-Preserving Output Restructuring for Ambient Visual Feedback in Speech Emotion Recognition: A Design-Oriented Comparative Study},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {100017--100035},
doi = {10.1109/access.2026.3706867}
}
EUVIMED