Skip to content

Dataset · disco-eth

EuroSpeech multilingual parliamentary speech

disco-eth/EuroSpeech

Large corpus of aligned parliamentary speech in 22 European languages for ASR and TTS.

What's inside

EuroSpeech is a large multilingual speech corpus collected from parliamentary proceedings and processed with an audio-to-non-verbatim-transcript alignment pipeline. It contains high-quality aligned speech in 22 European languages, with metadata listing de, en, mt, sl, sk, and da. Audio is provided at 16 kHz, with a 24 kHz version available. Volume ranges from 10 to 100 million examples.

What you can do with it

  • Training multilingual ASR models on parliamentary speech
  • Fine-tuning TTS systems for European languages
  • Testing ASR on non-verbatim transcripts and various audio formats
Tasks
Automatic speech recognition, Text → speech
Languages
German, English, Maltese, Slovenian, Slovak, Danish, Swedish, Norwegian and 4 more
Size
10M–100M rows
Downloads this month
87.8K
Likes
107

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("disco-eth/EuroSpeech")