Dataset · nvidia
Granary: speech and translation in 25 languages
nvidia/Granary
A large open speech dataset across 25 European languages for ASR and speech translation.
What's inside
Granary is a large open multilingual speech dataset for ASR and AST across 25 European languages. It unifies multiple collections into a single structure, addressing the shortage of high-quality data for low-resource languages. According to the dataset card, it spans roughly 1 million hours of speech, while metadata places its size at 100M<n<1B. Supported languages include bg, cs, da, de, el, en, and others.
What you can do with it
- Train a multilingual ASR model for European languages.
- Fine-tune a speech translation (AST) model.
- Evaluate ASR performance on low-resource languages.
- Tasks
- Automatic speech recognition, Translation
- Languages
- Bulgarian, Czech, Danish, German, Greek, English, Spanish, Estonian and 4 more
- Size
- 100M–1B rows
- Downloads this month
- 4.8K
- Likes
- 238
Models trained on it
- nvidia/parakeet-tdt-0.6b-v3Automatic speech recognition 557.5K
- nvidia/nemotron-3.5-asr-streaming-0.6bAutomatic speech recognition 417.4K
- FluidInference/parakeet-tdt-0.6b-v3-coremlAutomatic speech recognition 302.9K
- nvidia/canary-1b-v2Automatic speech recognition 227.7K
- nvidia/parakeet-tdt-0.6b-v2Automatic speech recognition 117.2K
- nvidia/nemotron-speech-streaming-en-0.6bAutomatic speech recognition 106.4K
- nvidia/parakeet-unified-en-0.6bAutomatic speech recognition 7.6K
- nvidia/nemotron-labs-unified-asr-0.6bAutomatic speech recognition 4.3K
How to load it
from datasets import load_dataset
ds = load_dataset("nvidia/Granary")