Dataset · UBC-NLP
Casablanca Arabic dialects recognition corpus
UBC-NLP/Casablanca
Audio recordings with transcriptions for training and testing ASR systems across various Arabic dialects.
What's inside
The dataset consists of Arabic audio-text pairs covering various dialects. The sample size ranges from 10 to 100 thousand examples. The project was created by the UBC-NLP group to address the scarcity of speech data for low-resource dialects.
What you can do with it
- Training and fine-tuning speech recognition (ASR) models for Arabic
- Evaluating robustness of speech models to different Arabic dialects
- Languages
- Arabic
- Size
- 10K–100K rows
- Downloads this month
- 2.1K
- Likes
- 39
Models trained on it
- touati-kamel/whisper-algerian-darja-smallAutomatic speech recognition 636
- Tilas/whisper-lv3-darija-marco-30hAutomatic speech recognition 106
- touati-kamel/whisper-algerian-darja-mediumAutomatic speech recognition 78
- deepdml/whisper-tiny-ar-mix-normAutomatic speech recognition 63
- deepdml/whisper-large-v3-turbo-ar-mix-normAutomatic speech recognition 62
- deepdml/whisper-small-ar-mix-normAutomatic speech recognition 59
- deepdml/whisper-base-ar-mix-normAutomatic speech recognition 52
- MohammedNasri/whisper-algerian-dialectAutomatic speech recognition 47
How to load it
from datasets import load_dataset
ds = load_dataset("UBC-NLP/Casablanca")