Omnilingual ASR Corpus (Subset)

This dataset is a subset of the original Omnilingual ASR Corpus by Facebook. It includes high-quality audio data for low-resource languages, designed for automatic speech recognition (ASR) tasks.

Key Change from Original:

• prompt_id → now id
• iso_639_3 → now audio_language
• raw_text → now text

---

Dataset Summary

Config	Language	Total Audios	Total Hours	Audios ≤30s	% ≤30s	Hours ≤30s
koo	Rukonjo	923	13.23	186	20.2%	1.07
cgg	Rukiga	805	10.84	191	23.7%	1.04
ttj	Rutooro	958	10.31	398	41.5%	2.01
pko	Pokot	465	10.43	15	3.2%	0.09

Total across all languages: 3,151 audios | ~44.81 hours

---

Splits Overview

koo – Rukonjo

Split	Audios	Hours	≤30s Audios	≤30s Hours
train	654	9.98	93 (14.2%)	0.57
dev	146	1.67	52 (35.6%)	0.30
test	123	1.57	41 (33.3%)	0.20

cgg – Rukiga

Split	Audios	Hours	≤30s Audios	≤30s Hours
train	559	7.77	118 (21.1%)	0.62
dev	126	1.50	38 (30.2%)	0.22
test	120	1.57	35 (29.2%)	0.21

ttj – Rutooro

Split	Audios	Hours	≤30s Audios	≤30s Hours
train	674	7.03	287 (42.6%)	1.49
dev	110	1.61	18 (16.4%)	0.11
test	174	1.66	93 (53.4%)	0.41

pko – Pokot

Split	Audios	Hours	≤30s Audios	≤30s Hours
train	298	6.74	10 (3.4%)	0.06
dev	70	1.63	2 (2.9%)	0.01
test	97	2.05	3 (3.1%)	0.02

---

How to Load the Dataset

from datasets import load_dataset

# Load a specific language config
dataset = load_dataset("evie-8/omnilingual", "koo")

# Access splits
train = dataset["train"]
dev   = dataset["dev"]
test  = dataset["test"]

# Example: View first entry
print(train[0])

Available Configurations

• "koo" – Rukonzo
• "cgg" – Rukiga
• "ttj" – Rutooro
• "pko" – Pokot

Special tags

The following special tags were used in transcriptions (text field) to mark laughter, fillers and other types of non-verbal content:

Tag	Purpose
<laugh>	The sound of laughter.
<hesitation>	A hesitation sound, often used by speakers while thinking of the next thing to say. In English, some common hesitation sounds are “err”, “um”, “huh”, etc.
<unintelligible>	A word or sequence of words that cannot be understood.
<noise>	Any other type of noise, such as the speaker coughing or clearing their throat, a car honking, the sound of something hitting the microphone, a phone buzzing, etc.

Data Fields

Field	Type	Description
id	string	Unique identifier (formerly prompt_id)
audio	Audio()	Audio data
audio_language	string	Language code of the audio (e.g., koo)
text	string	Transcription
prompt	string	Original prompt or instruction
duration	float64	Duration of the audio in seconds
speaker_id	string	Identifier for the speaker

Inspect with:

print(dataset["train"].features)

---

Access Requirements

You must agree to share your contact information to access this dataset on Hugging Face.

Accept Terms & Access Dataset

---

Citation

@misc{omnilingualasr2025,
    title={{Omnilingual ASR}: Open-Source Multilingual Speech Recognition for 1600+ Languages},
    author={{Omnilingual ASR Team} and Keren, Gil and Kozhevnikov, Artyom and Meng, Yen and Ropers, Christophe and Setzler, Matthew and Wang, Skyler and Adebara, Ife and Auli, Michael and Chan, Kevin and Cheng, Chierh and Chuang, Joe and Droof, Caley and Duppenthaler, Mark and Duquenne, Paul-Ambroise and Erben, Alexander and Gao, Cynthia and Mejia Gonzalez, Gabriel and Lyu, Kehan and Miglani, Sagar and Pratap, Vineel and Sadagopan, Kaushik Ram and Saleem, Safiyyah and Turkatenko, Arina and Ventayol-Boada, Albert and Yong, Zheng-Xin and Chung, Yu-An and Maillard, Jean and Moritz, Rashel and Mourachko, Alexandre and Williamson, Mary and Yates, Shireen},
    year={2025},
    url={https://ai.meta.com/research/publications/omnilingual-asr-open-source-multilingual-speech-recognition-for-1600-languages/},
}

---

Original Source

facebook/omnilingual-asr-corpus

---