ghana-speech-eval

⚠️ Evaluation only. This is a held-out benchmark for measuring ASR quality on Ghanaian languages. Please do not train or fine-tune on it — doing so contaminates the benchmark and makes reported scores meaningless. Every subset ships a single eval split.

Multi-source ASR evaluation benchmark for Ghanaian languages. All subsets are 16 kHz mono with standardised schema (audio, text, language, country, length, iso, subset).

Source groups

Group	Description	Clip length	Per-language cap
jw_*	JW.org speech (AfriSpeech v1)	3–15 s	1 000
bible_*	Bible-audio aligned (ghana-speech)	3–15 s	1 000
unicef_*	UNICEF health recordings	3–60 s	all available (~195–200)
finance_*	Finance-domain speech	0–60 s	1 000

Twi note: bible_twi_twi merges Asante and Akuapem Twi (500/500 balanced). finance_twi also merges Asante and Akuapem Twi (500/500 balanced).

Configs (57 total)

JW group (jw_*) — 10 configs

Config	ISO	Language
jw_ahanta_aha	aha	Ahanta
jw_dagaare_dga	dga	Dagaare
jw_dangme_ada	ada	Dangme
jw_ewe_ewe	ewe	Ewe
jw_fante_fat	fat	Fante
jw_frafra_gur	gur	Frafra
jw_ga_gaa	gaa	Ga
jw_nzema_nzi	nzi	Nzema
jw_sehwi_sfw	sfw	Sehwi
jw_twi_twi	twi	Twi

Bible group (bible_*) — 41 configs

Config	ISO	Language
bible_twi_twi	twi	Twi
bible_Fante_fat	fat	Fante
bible_Ewe_ewe	ewe	Ewe
bible_Dagbani_dag	dag	Dagbani
bible_Dangme_ada	ada	Dangme
bible_Nzema_nzi	nzi	Nzema
bible_Sehwi_sfw	sfw	Sehwi
bible_Kusaal_kus	kus	Kusaal
bible_Mampruli_maw	maw	Mampruli
bible_Fulfulde_Maasina_ffm	ffm	Fulfulde
bible_Bissa_bib	bib	Bissa
bible_Hausa_hau	hau	Hausa
bible_Kabiye_kbp	kbp	Kabiye
bible_Nawuri_naw	naw	Nawuri
bible_Konkomba_xon	xon	Konkomba
bible_Gonja_gjn	gjn	Gonja
bible_Kasem_xsm	xsm	Kasem
bible_Konni_kma	kma	Konni
bible_Ntrubo_ntr	ntr	Ntrubo
bible_Bimoba_bim	bim	Bimoba
bible_Buli_bwu	bwu	Buli
bible_Chumburung_ncu	ncu	Chumburung
bible_Dagaare_dga	dga	Dagaare
bible_Deg_mzw	mzw	Deg
bible_Gikyode_acd	acd	Gikyode
bible_Lelemi_lef	lef	Lelemi
bible_Nkonya_nko	nko	Nkonya
bible_Paasaal_sig	sig	Paasaal
bible_Sekpele_lip	lip	Sekpele
bible_Selee_snw	snw	Selee
bible_Sisaala_Tumulung_sil	sil	Sisaala
bible_Siwu_akp	akp	Siwu
bible_Tampulma_tpm	tpm	Tampulma
bible_Tem_kdh	kdh	Tem
bible_Tuwuli_bov	bov	Tuwuli
bible_Vagla_vag	vag	Vagla
bible_Anyin_any	any	Anyin
bible_Avatime_avn	avn	Avatime
bible_Bassar_Ntcham_bud	bud	Bassar
bible_Birifor_Southern_biv	biv	Birifor
bible_Ninkare_gur	gur	Ninkare

UNICEF group (unicef_*) — 3 configs

Config	ISO	Language
unicef_ewe	ewe	Ewe
unicef_dagbani	dag	Dagbani
unicef_twi	twi	Twi

Finance group (finance_*) — 3 configs

Config	ISO	Language	Notes
finance_ga	gaa	Ga
finance_fante	fat	Fante
finance_twi	twi	Twi	500 Asante + 500 Akuapem

Loading

from datasets import load_dataset

# Load a specific subset (config); every subset has a single `eval` split
ds = load_dataset("ghananlpcommunity/ghana-speech-eval", "finance_twi", split="eval")

# Inspect available subsets
from datasets import get_dataset_config_names
print(get_dataset_config_names("ghananlpcommunity/ghana-speech-eval"))

Schema

Column	Type	Description
audio	Audio	Decoded waveform (16 kHz mono)
text	string	Reference transcription
language	string	Language label
country	string	ISO 3166-1 alpha-2
length	float64	Clip duration (seconds)
iso	string	ISO 639-3 code
subset	string	<prefix>_<config>

License

CC-BY-4.0. Respect the terms of each source dataset.