# WAXAL: multilingual African language speech corpus (ASR and TTS)

## Facts

- Provider: Google, with Makerere University, University of Ghana, Digital Umuganda, Media Trust, Loud and Clear, AIMS Senegal
- License: CC-BY-SA-4.0 (https://huggingface.co/datasets/google/WaxalNLP)
- Size: 2,619 files
- Format: JSON, Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/google/WaxalNLP
- Access: public, readable without an account
- Category: Speech & Audio
- Source: https://huggingface.co/papers/2602.02734
- Page: https://scigantic.com/archive/8b8134f8-3490-5b08-8f03-40f5dfcea1a1
- Keywords: Speech & Audio, speech recognition, text-to-speech, JSON, Parquet, Hugging Face

## About

WAXAL is a large-scale multilingual speech corpus for African languages, released by Google with partner institutions and funded by Google and the Gates Foundation under an agreement to make it openly accessible. The ASR part covers 19 African languages with approximately 1,250 hours of transcribed natural speech from many voices, representing over 100 million speakers across 40 Sub-Saharan countries (including Acholi, Luganda, Akan, Ewe, Dagbani, Fula, Lingala, Shona, Malagasy, Amharic, Oromo, Tigrinya and Wolaytta). The TTS part covers 17 languages with over 180 hours of high-quality single-speaker recordings of phonetically balanced scripts (including Hausa, Igbo, Yoruba, Nigerian Pidgin, Kikuyu, Luo, Swahili, Bambara and Wolof). This repository holds data/ASR (19 language folders), data/ASR_v2 (speaker-disjoint splits) and data/TTS (18 folders) as Parquet shards, plus a Croissant metadata file (measured 2026-11-14; data/ASR/aka alone is 278 shards, about 57 GB; the Hub reports about 821 GB). Load with the datasets library per language. Licences are set per provider and are all open: University of Ghana data CC BY 4.0, all other providers CC BY-SA 4.0; keep the per-language licence when redistributing.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("google/WaxalNLP")
print(ds)
```

## Data structure

File types in sampled files: JSON, Parquet

Example file: `data/ASR/ach/ach-test-00000.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/8b8134f8-3490-5b08-8f03-40f5dfcea1a1
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/8b8134f8-3490-5b08-8f03-40f5dfcea1a1
- Over MCP: `get_schema_card` with id `8b8134f8-3490-5b08-8f03-40f5dfcea1a1` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [JSON datasets](https://scigantic.com/topics/json)
- [Speech & Audio datasets](https://scigantic.com/topics/speech-and-audio)
- [Speech recognition datasets](https://scigantic.com/topics/speech-recognition)
- [Text-to-speech datasets](https://scigantic.com/topics/text-to-speech)
