Skip to content

Dataset · speechcolab

39,000 Hours of Arabic Speech Recognition Audio

speechcolab/gigaspeech3

Large weakly-supervised ASR corpus covering five Arabic dialects.

What's inside

GigaSpeech 3 is a large-scale Arabic speech recognition (ASR) audio dataset containing about 10 million segments with a total duration of 39,000 hours. The recordings feature weak transcriptions and cover five dialects: Algerian, Egyptian, Iraqi, Saudi, and Syrian.

What you can do with it

  • Training multilingual and dialectal Arabic speech recognition (ASR) models
  • Researching weak supervision learning methods for audio
  • Adapting and fine-tuning models for low-resource Arabic dialects
Tasks
Automatic speech recognition
Languages
Arabic
Size
10M–100M rows
Downloads this month
24.2K
Likes
33

How to load it

from datasets import load_dataset

ds = load_dataset("speechcolab/gigaspeech3")