goobolabs/somnlp-stt-corpus
Transcribed Somali speech with word-level alignment, sourced from radio and read passages, for ASR training.
View on Hugging FaceModality
Speech
Size
320 hours · 48 GB
License
CC-BY-4.0
Language
Somali (af)
Format
MP3 + Parquet
Version
v1.2
Tasks
ASR, Speaker ID, VAD
Domains
Radio · Read speech · Interviews
| Split | Rows | Size |
|---|---|---|
| train | 12,800 clips | 43.1 GB |
| dev | 640 clips | 2.4 GB |
| test | 640 clips | 2.5 GB |
from datasets import load_dataset
ds = load_dataset("goobolabs/somnlp-stt-corpus", "clean")
sample = ds["train"][0]
print(sample)@dataset{somnlp_stt_2026,
title = {SomNLP-STT-Corpus: Somali Speech Dataset},
author = {Goobo Labs},
year = {2026},
url = {https://huggingface.co/datasets/goobolabs/somnlp-stt-corpus}
}