---
license: cc-by-4.0
language:
- sw
task_categories:
- automatic-speech-recognition
---

# Afrivoice Swahili — All Domains

Combined dataset across all 5 domains from the original source.

## Attribution

**Original dataset:** [DigitalUmuganda/Afrivoice_Swahili](https://huggingface.co/datasets/DigitalUmuganda/Afrivoice_Swahili)  
**License:** CC-BY-4.0  
**Attribution:** DigitalUmuganda  

This dataset is derived from the above source and released under the same CC-BY-4.0 license.

## Domains
- Agriculture (~124k train samples)
- Education (~69k train samples)
- Financial (~104k train samples)
- Government (~99k train samples)
- Health (~118k train samples)

## Splits
- `train`: all 5 domain train splits combined (~514k samples)
- `validation`: all 5 domain dev splits combined
- `test`: all 5 domain test splits combined

## Features
- `audio`: 16kHz mono float32 array
- `transcription`: normalized Swahili text
- `domain`: source domain label