Dataset · FBK-MT
Multilingual spoken language understanding corpus
FBK-MT/Speech-MASSIVE
Spoken version of the MASSIVE text corpus for ASR, intent detection, and slot filling tasks.
What's inside
The dataset is an audio version of part of the MASSIVE text corpus, retaining intent prediction and slot-filling annotations. It covers 12 languages from various language families, including Russian, English, Arabic, German, Korean, and others. The total size ranges from 10,000 to 100,000 records.
What you can do with it
- Training multilingual spoken language understanding (SLU) models and voice assistants
- Training ASR followed by intent classification
- Zero-shot audio classification across multiple languages
- Tasks
- Audio classification, Text classification, Zero shot classification, Automatic speech recognition
- Languages
- Arabic, German, Spanish, French, Hungarian, Korean, Dutch, Polish and 4 more
- Size
- 10K–100K rows
- Downloads this month
- 3.6K
- Likes
- 57
Models trained on it
- Trendency/whisper-large-v3-huAutomatic speech recognition 628
- naver/mHuBERT-147-ASR-frAutomatic speech recognition 60
- gs224/whisper-tiny-polishAutomatic speech recognition 53
- alkiskoudounas/xls-r-128-speechmassive-de-DEAudio classification 40
- alkiskoudounas/xls-r-53-de-speechmassive-de-DEAudio classification 31
- alkiskoudounas/xls-r-128-speechmassive-fr-FRAudio classification 26
- alkiskoudounas/xls-r-53-fr-speechmassive-fr-FRAudio classification 24
- alkiskoudounas/xls-r-128-speechmassive-fr-FR-goldAudio classification 17
How to load it
from datasets import load_dataset
ds = load_dataset("FBK-MT/Speech-MASSIVE")