Open dataset · CC-BY-4.0
A speech corpus for Ethiopian languages. Recorded by speakers who gave informed consent and were paid for the work, verified before release, and published for anyone to use.
51 downloads on Hugging Face in the last 30 days.
Releases
Speech to text · Read speech
tir · v2605 · released 2026-05-06
Every speaker gave informed consent to this public release and was paid wages for the work. Most speakers live and work in Tigray. Transcriptions are in Ge'ez script, verified against the audio before release.
Tigrinya · v2605
| Split | Hours | Segments | Speakers |
|---|---|---|---|
| train | 11.7 | 3,393 | 137 |
| dev | 1.5 | 398 | 12 |
| test | 1.5 | 387 | 12 |
Splits are speaker-disjoint and gender-stratified: twelve speakers each in dev and test, six male and six female, with no speaker overlap into training. The test split is frozen and will not change in future versions, so results reported on it are comparable to anyone else's.
Audio, transcription in Ge'ez script, duration, anonymized speaker ID, gender, word count, speaking rate, and the version it was introduced in.
WAV, 16 kHz, 16-bit, mono. 1.68 GB. Read speech.
Training and evaluating Tigrinya speech recognition, cross-lingual research, linguistic study and benchmarking. Not for speaker identification or surveillance.
Collection
Before a contributor recorded a segment, we met with them over video and explained what the data would be used for and how it would be released. Every speaker gave informed consent to a public CC-BY-4.0 release.
Contributors were paid wages for the work. The compensation structure was explained to each of them in the same meeting, before they recorded anything.
Contributors read to us live and were coached on pacing, clarity and consistency. As recordings came in, we reviewed quality, gave feedback and helped them improve.
Use it
from datasets import load_dataset
ds = load_dataset("phoneticoai/phonetico-speech", "tir", split="train")
# Pin to a version
ds = load_dataset("phoneticoai/phonetico-speech", "tir", revision="v2605")Versions are sequential. Future versions expand existing languages and add new ones; every segment carries the version it was introduced in.
@misc{phonetico-speech,
title = {Phonetico Speech: A Speech Corpus for Ethiopian Language ASR},
author = {Feyissa, Solen},
year = {2026},
url = {https://huggingface.co/datasets/phoneticoai/phonetico-speech},
note = {v2605}
}Licence: CC-BY-4.0. Questions and removal requests: [email protected]. Announcement: the release post.