Europarl-ASR/README.md

Europarl-ASR

Europarl-ASR: A Large Speech+Text Corpus of Parliamentary Debates for Streaming ASR Benchmarking and Speech Data Filtering/Verbatimization.

• 1300 hours EN transcribed speech data.
• 18 hours EN speech data w/ revised verbatim and official non-verbatim transcriptions, split in 2 dev/test partitions for 2 realistic ASR tasks.
• 3 full sets of timed transcriptions for the training data: official non-verbatim, automatically noise-filtered, and automatically verbatimized.
• 70 million tokens EN text data.