Datasetpublic

POTOMITAN/luxembourgish-corpus

CorpusLux: Luxembourgish Speech Transcriptions Dataset of Luxembourgish speech transcriptions from government press conferences (2025-2026), formatted as speaker-text pairs (inspired by POTOMITAN/PawolKreyol-gfc). Dataset Structure Data Splits Train: 125 turns (80%) Validation: 16 turns (10%) Test: 16 turns (10%) Total: 157 turns from 6 documents Features Feature Type Description Source string Speaker name (e.g., "Gilles… See the full description on the dataset page: https://huggingface.co/datasets/POTOMITAN/luxembourgish-corpus.

sourceHugging Faceunknownupdated 1mo agoView on Hugging Face

1likes34downloads

Dataset Card

CorpusLux: Luxembourgish Speech Transcriptions

Dataset of Luxembourgish speech transcriptions from government press conferences (2025-2026), formatted as speaker-text pairs (inspired by POTOMITAN/PawolKreyol-gfc).

Dataset Structure

Data Splits

—Train: 125 turns (80%)
—Validation: 16 turns (10%)
—Test: 16 turns (10%)

Total: 157 turns from 6 documents

Features

Feature	Type	Description
`Source`	string	Speaker name (e.g., "Gilles Roth", "Claude Meisch")
`Texte`	string	Spoken text in Luxembourgish

Usage

python

from datasets import load_dataset

# Depuis Hugging Face Hub
dataset = load_dataset("username/corpus_lux")

# Depuis local
dataset = load_dataset("output/corpus_lux")

# Accéder aux données
train_data = dataset["train"]
for example in train_data.select(range(5)):
    print(f"Source: {example['Source']}")
    print(f"Texte: {example['Texte'][:100]}...")

Dataset Statistics

—Total turns: 157
—Unique speakers: 10
—Language: Luxembourgish (lb)
—Period: 2025-2026
—Source: Luxembourg Government press briefings

Notes

This dataset contains transcriptions of Luxembourgish government press conferences. Each example represents a single speaking turn (speaker + their spoken text).

The text is in Luxembourgish with some French and German passages.