ghanaopenai/ghana-speech-female-22khz_bible

Ghana Speech - Female 22kHz Voice Converted (Scripture / Bible)

This dataset contains 499.21 hours of natural Ghanaian speech converted into a high-fidelity African female voice at 22.05 kHz, derived from scripture and Bible recordings in ghananlpcommunity/ghana-speech.

• Total Scale: 499.21 hours across 42 Ghanaian language subsets (235,613 clips, 35.06 GB).
• Target Voice: clear-high-slow (native Twi female voice, pitch ~261 Hz, clear timbre, lively intonation) from AfriSpeech/africa-female-vc.
• Audio Format: 22,050 Hz (22.05 kHz) mono 16-bit FLAC.
• Model: Seed-VC (seed-uvit-whisper-small-wavenet) + NVIDIA BigVGAN 22kHz neural vocoder.
• Viewer-Ready: Shards are at most ~175 MB, with page index enabled (write_page_index=True, row_group_size=500) for instant random-access playback in the Hugging Face Dataset Viewer.

---

Active Language Subsets

Language	Subset (Config)	ISO 639-3	Segments	Duration	Shards	File Size
Akuapem Twi	Akuapem_Twi_twi	twi	7,292	10.00h	5	699.7 MB
Anyin	Anyin_any	any	7,260	20.00h	7	1430.6 MB
Asante Twi	Asante_Twi_twi	twi	11,506	20.01h	11	1456.6 MB
Avatime	Avatime_avn	avn	9,259	20.00h	7	1398.7 MB
Bassar (Ntcham)	Bassar_Ntcham_bud	bud	11,090	20.00h	7	1275.1 MB
Bimoba	Bimoba_bim	bim	7,814	20.00h	5	1316.1 MB
Southern Birifor	Birifor_Southern_biv	biv	7,942	18.82h	5	1191.2 MB
Bissa	Bissa_bib	bib	8,799	19.98h	6	1309.5 MB
Buli	Buli_bwu	bwu	6,192	14.11h	5	869.1 MB
Chumburung	Chumburung_ncu	ncu	3,897	10.00h	5	654.4 MB
Dagaare	Dagaare_dga	dga	4,697	9.93h	5	684.2 MB
Dagbani	Dagbani_dag	dag	4,355	10.00h	5	671.4 MB
Dangme	Dangme_ada	ada	4,379	10.00h	5	663.4 MB
Deg	Deg_mzw	mzw	4,960	10.00h	5	651.2 MB
Ewe	Ewe_ewe	ewe	4,720	10.01h	5	673.3 MB
Fante	Fante_fat	fat	4,147	10.02h	4	628.6 MB
Fulfulde (Maasina)	Fulfulde_Maasina_ffm	ffm	5,745	10.01h	5	698.7 MB
Gikyode	Gikyode_acd	acd	3,843	10.01h	5	671.4 MB
Gonja	Gonja_gjn	gjn	4,607	10.00h	5	686.6 MB
Hausa	Hausa_hau	hau	4,887	10.00h	4	635.8 MB
Kabiye	Kabiye_kbp	kbp	5,887	10.01h	5	644.7 MB
Kasem	Kasem_xsm	xsm	5,195	10.00h	5	671.7 MB
Konkomba	Konkomba_xon	xon	5,065	10.00h	4	608.5 MB
Konni	Konni_kma	kma	4,110	10.01h	5	645.2 MB
Kusaal	Kusaal_kus	kus	5,066	10.00h	5	691.0 MB
Lelemi	Lelemi_lef	lef	4,867	10.00h	5	664.8 MB
Mampruli	Mampruli_maw	maw	3,847	10.00h	5	672.4 MB
Nawuri	Nawuri_naw	naw	4,789	10.00h	5	685.6 MB
Ninkare (Frafra / Gurene)	Ninkare_gur	gur	4,946	10.00h	5	668.6 MB
Nkonya	Nkonya_nko	nko	6,103	10.01h	5	704.4 MB
Ntrubo	Ntrubo_ntr	ntr	5,423	10.01h	5	662.0 MB
Nzema	Nzema_nzi	nzi	5,120	10.01h	5	723.8 MB
Paasaal	Paasaal_sig	sig	3,846	10.01h	4	640.3 MB
Sehwi	Sehwi_sfw	sfw	7,704	16.20h	8	1129.1 MB
Sekpele	Sekpele_lip	lip	3,809	10.00h	5	652.8 MB
Selee	Selee_snw	snw	3,875	10.00h	5	658.6 MB
Tumulung Sisaala	Sisaala_Tumulung_sil	sil	3,822	10.00h	5	669.6 MB
Siwu	Siwu_akp	akp	5,543	10.01h	5	703.2 MB
Tampulma	Tampulma_tpm	tpm	4,120	10.01h	5	656.1 MB
Tem	Tem_kdh	kdh	4,710	10.02h	5	656.3 MB
Tuwuli	Tuwuli_bov	bov	5,315	10.01h	5	704.4 MB
Vagla	Vagla_vag	vag	5,060	10.00h	5	657.2 MB
Total	42 languages	—	235,613	499.21h	222	35.06 GB

---

How to Load in Python

from datasets import load_dataset

# Load Akan (Asante Twi) female speech (22.05 kHz)
ds = load_dataset("ghanaopenai/ghana-speech-female-22khz_bible", "Asante_Twi_twi", split="train")

# Load Ewe female speech
ds_ewe = load_dataset("ghanaopenai/ghana-speech-female-22khz_bible", "Ewe_ewe", split="train")

sample = ds[0]
print(sample["id"], sample["text"], sample["duration"])
# Play audio
# sample["audio"]["array"], sample["audio"]["sampling_rate"] (22050)

---

License & Attribution

Shared under CC BY-NC 4.0 (Attribution-NonCommercial). Original recordings from Street is Watching / Ghana NLP Community scripture corpus.