kapturecx/ohun

ohùn — Igbo · Yorùbá · Hausa · Pidgin speech corpus

ohùn (Yorùbá for voice) merges the three WaZoBiaSpeech corpora published by Africanvoice into a single repository, so all three of Nigeria's major languages can be pulled from one place.

Audio is byte-identical to the sources — this repo re-registers the very same objects, it does not re-encode anything.

Contents

718,336 utterances · 1,035 GB of audio across four languages.

config	split	rows	shards	size
ibo	train	127,193	4,996	139.4 GB
ibo	dev	5,569	41	6.8 GB
ibo	dev_test	6,945	86	10.8 GB
yor	train	147,686	408	204.0 GB
yor	dev	7,717	25	12.0 GB
yor	dev_test	8,882	26	12.5 GB
hau	train	40,115	337	168.1 GB
hau	dev	2,253	21	10.3 GB
hau	dev_test	1,627	20	9.5 GB
pcm	train	335,288	839	419.3 GB
pcm	dev	17,737	40	19.6 GB
pcm	dev_test	17,324	46	22.7 GB

Usage

from datasets import load_dataset

ibo = load_dataset("kapturecx/ohun", "ibo", split="train", streaming=True)  # Igbo
yor = load_dataset("kapturecx/ohun", "yor", split="train", streaming=True)  # Yorùbá
hau = load_dataset("kapturecx/ohun", "hau", split="train", streaming=True)  # Hausa
pcm = load_dataset("kapturecx/ohun", "pcm", split="train", streaming=True)  # Nigerian Pidgin

Each language is a separate config because the three sources do not share a schema — see below. Streaming is recommended; the full corpus is ~1.0 TB.

Schemas

The source corpora differ in their column sets, so they are kept as-is rather than being lossily forced into a common table:

ibo

column	type
audio_id	string
speaker_id	string
audio_path	struct<bytes: binary, path: string>
transcript	string
duration_seconds	double
gender	string
age_group	string
education	string
domain	string
type	string
split	string
language	string
rel_audio_path	string

yor

column	type
transcript	string
audio_id	string
speaker_id	string
gender	string
age_group	string
education	string
domain	string
type	string
split	string
language	string
audio_path	struct<bytes: binary, path: string>

hau

column	type
audio_id	string
speaker_id	string
audio_path	string
transcript	string
duration_seconds	double
gender	string
age_group	string
education	string
domain	string
type	string
split	string
audio	struct<bytes: binary, path: string>

pcm

column	type
transcript	string
audio_id	string
speaker_id	string
gender	string
age_group	string
education	string
domain	string
type	string
split	string
language	string
audio_path	struct<bytes: binary, path: string>

• audio_path differs in type: ibo is struct<bytes: binary, path: string>, yor is struct<bytes: binary, path: string>, hau is string, pcm is struct<bytes: binary, path: string>
• duration_seconds is present only in ibo, hau
• language is present only in ibo, yor, pcm
• rel_audio_path is present only in ibo
• audio is present only in hau
• yor, pcm share an identical schema

Sources & license

config	source dataset
ibo	Africanvoice/African_voices_igbo
yor	Africanvoice/African_voices_yoruba
hau	Africanvoice/African_voices_hausa
pcm	Africanvoice/African_voices_naija

Licensed CC-BY-4.0, following the upstream corpora. Please cite the original WaZoBiaSpeech datasets. The upstream READMEs note these corpora were still work-in-progress at the time of the snapshot merged here.