Dataset · facebook
Meta ASR corpus in 348 rare languages
facebook/omnilingual-asr-corpus
Spontaneous speech recordings and transcriptions for 348 languages for ASR training and language identification.
What's inside
This is a collection of spontaneous speech recordings and their transcriptions for 348 underrepresented languages. The corpus was collected as part of the Meta FAIR Omnilingual ASR project for ASR training and spoken language identification. The schema includes language, iso_639_3, and iso_15924 fields; modalities are audio and text.
What you can do with it
- Train or fine-tune an ASR model for rare languages
- Evaluate spoken language identification
- Create a LoRA for a low-resource language
- Tasks
- Automatic speech recognition, Audio classification
- Languages
- aae, aal, aao, abn, abr, abs, abv, acm and 4 more
- Size
- 100K–1M rows
- Downloads this month
- 9.5K
- Likes
- 222
Models trained on it
- SWivid/Habibi-TTS 44.5K
- third-intelligence/RubatoCodecAudio → audio 723
- bezzam/omniasr-ctc-300m-v2Automatic speech recognition 391
- bezzam/omniasr-llm-300m-v2Automatic speech recognition 380
- bezzam/omniasr-ctc-1b-v2Automatic speech recognition 291
- mohamedalgmaal/Habibi-TTS 229
- bezzam/omniasr-llm-1b-v2Automatic speech recognition 212
- bezzam/omniasr-llm-7b-v2Automatic speech recognition 201
How to load it
from datasets import load_dataset
ds = load_dataset("facebook/omnilingual-asr-corpus")