Learning More with Less: Self-Supervised Approaches for Low-Resource Speech Emotion Recognition

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Gong, Ziwei, Shi, Pengyuan, Donbekci, Kaan, Ai, Lin, Chen, Run, Sasu, David, Wu, Zehui, Hirschberg, Julia
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913871963357184
author Gong, Ziwei
Shi, Pengyuan
Donbekci, Kaan
Ai, Lin
Chen, Run
Sasu, David
Wu, Zehui
Hirschberg, Julia
author_facet Gong, Ziwei
Shi, Pengyuan
Donbekci, Kaan
Ai, Lin
Chen, Run
Sasu, David
Wu, Zehui
Hirschberg, Julia
contents Speech Emotion Recognition (SER) has seen significant progress with deep learning, yet remains challenging for Low-Resource Languages (LRLs) due to the scarcity of annotated data. In this work, we explore unsupervised learning to improve SER in low-resource settings. Specifically, we investigate contrastive learning (CL) and Bootstrap Your Own Latent (BYOL) as self-supervised approaches to enhance cross-lingual generalization. Our methods achieve notable F1 score improvements of 10.6% in Urdu, 15.2% in German, and 13.9% in Bangla, demonstrating their effectiveness in LRLs. Additionally, we analyze model behavior to provide insights on key factors influencing performance across languages, and also highlighting challenges in low-resource SER. This work provides a foundation for developing more inclusive, explainable, and robust emotion recognition systems for underrepresented languages.
format Preprint
id arxiv_https___arxiv_org_abs_2506_02059
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Learning More with Less: Self-Supervised Approaches for Low-Resource Speech Emotion Recognition
Gong, Ziwei
Shi, Pengyuan
Donbekci, Kaan
Ai, Lin
Chen, Run
Sasu, David
Wu, Zehui
Hirschberg, Julia
Sound
Computation and Language
Speech Emotion Recognition (SER) has seen significant progress with deep learning, yet remains challenging for Low-Resource Languages (LRLs) due to the scarcity of annotated data. In this work, we explore unsupervised learning to improve SER in low-resource settings. Specifically, we investigate contrastive learning (CL) and Bootstrap Your Own Latent (BYOL) as self-supervised approaches to enhance cross-lingual generalization. Our methods achieve notable F1 score improvements of 10.6% in Urdu, 15.2% in German, and 13.9% in Bangla, demonstrating their effectiveness in LRLs. Additionally, we analyze model behavior to provide insights on key factors influencing performance across languages, and also highlighting challenges in low-resource SER. This work provides a foundation for developing more inclusive, explainable, and robust emotion recognition systems for underrepresented languages.
title Learning More with Less: Self-Supervised Approaches for Low-Resource Speech Emotion Recognition
topic Sound
Computation and Language
url https://arxiv.org/abs/2506.02059