LuxEmo: Expressive Text-to-Speech Corpus for Luxembourgish

2026-06-30

AI Summary

This study addresses the longstanding scarcity of high-quality, expressive speech data for low-resource languages by presenting LuxEmo, the first 21-hour Luxembourgish conversational emotional speech corpus, encompassing four emotion categories and sourced from RTL youth radio programs. The authors propose an efficient semi-automated construction pipeline that integrates voice activity detection, denoising, language identification, LuxASR-based segmentation, automatic emotion prediction, lexical cues, and human verification. Through both objective metrics and subjective human evaluation, the corpus demonstrates strong efficacy in supporting expressive speech synthesis systems. This work thus provides a valuable resource and methodological framework for emotional speech research in under-resourced languages.

Abstract

State-of-the-art speech datasets predominantly focus on widely spoken languages, often overlooking low-resource languages such as Luxembourgish, which remain underrepresented in speech technology research. In this work, we introduce LuxEmo, a 21-hour conversational expressive speech corpus for Luxembourgish with 4 emotion categories. LuxEmo is derived from Radio Télévision Luxembourg (RTL) youth broadcasts, using automated detection followed by human validation. We propose a semi-automatic curation workflow combining voice activity detection, denoising, language identification, LuxASR-based segmentation, automatic emotion prediction, lexical cues, and targeted human review. Additionally, we benchmark five expressive TTS systems covering German-based cross-lingual transfer, multilingual Luxembourgish support, Luxembourgish adaptation, and non-parametric prosody transfer. Performance is evaluated using both objective metrics and human evaluation.

Problem

• low-resource languages
• Luxembourgish
• expressive speech corpus
• text-to-speech
• speech technology

Innovation

• expressive speech corpus
• low-resource language
• semi-automatic curation
• emotion-aware TTS
• Luxembourgish