Skip to content

Dataset · fixie-ai

CoVoST2 Subset with Embedded Audio

fixie-ai/covost2

Validation and test splits of the CoVoST2 speech translation dataset with embedded audio files.

What's inside

A repackaged version of the famous CoVoST2 speech translation dataset with embedded audio files for convenient viewing in the Hugging Face interface. The set includes only validation and test splits (ranging from 1 to 10 million entries), with only French, Spanish, and Chinese kept for translation into English, and audio duplicated for directions from English.

What you can do with it

  • Evaluating speech-to-text translation model quality
  • Quick viewing and debugging of audio-text pipelines without downloading separate archives
Size
1M–10M rows
Downloads this month
21.7K
Likes
8

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("fixie-ai/covost2")