metadata
language:
- so
task_categories:
- automatic-speech-recognition
pretty_name: Somali Domain-Specific ASR Dataset
Somali Domain-Specific ASR Dataset
This dataset contains Somali speech recordings and corresponding transcriptions collected in two domains: Agriculture and Health.
Dataset Structure
- Language: Somali
- Total recordings: 90,412
- Total duration: Approximately 154.6 hours
- Domains: Agriculture (50,798) and Health (39,614)
- Split: Train only
- Original columns: 23
- Format: Apache Arrow with embedded audio
Data Organization
The dataset contains ten Arrow batches (batch_0000 to batch_0009) stored under processed_155h/.
Each batch preserves the audio recordings, transcriptions, and associated metadata.