Datasetpublic

multilingual-tts/open-bible

OpenBibleTTS OpenBibleTTS is a large-scale, multilingual speech corpus for low-resource text-to-speech (TTS), spanning 37 underrepresented languages across five regions. It contains ~3,469 hours of aligned, verse-level read speech and 1,121,956 utterances, derived from the Open Bible platform and released under a permissive license. Alignment pipeline: https://github.com/davidguzmanr/open-bible-resources Source: Open Bible (CC BY-SA) Languages Africa (19), South… See the full description on the dataset page: https://huggingface.co/datasets/multilingual-tts/open-bible.

sourceHugging Facecc-by-sa-4.0updated 3mo agoView on Hugging Face

1likes2.8kdownloads

Dataset Card

OpenBibleTTS

OpenBibleTTS is a large-scale, multilingual speech corpus for low-resource text-to-speech (TTS), spanning 37 underrepresented languages across five regions. It contains ~3,469 hours of aligned, verse-level read speech and 1,121,956 utterances, derived from the Open Bible platform and released under a permissive license.

—Alignment pipeline: https://github.com/davidguzmanr/open-bible-resources
—Source: Open Bible (CC BY-SA)

Languages

Africa (19), South Asia (13), Southeast Asia (2), Middle East (2), Caribbean (1); nine language families and seven scripts. Load each language by its config name (first column). Note: Arabic Standard = Egyptian Arabic (arb) and Ndebele = Northern Ndebele (nde).

Config name	ISO 639-3	Region	Family	Script	Hours	Utterances
Arabic Standard	arb	Africa	Afro-Asiatic	Arabic	85.23	30,510
Assamese	asm	South Asia	Indo-European	Bengali	104.49	30,530
Bengali	ben	South Asia	Indo-European	Bengali	98.02	30,535
Central Kurdish	ckb	Middle East	Indo-European	Arabic	85.40	30,565
Chhattisgarhi	hne	South Asia	Indo-European	Devanagari	101.61	30,451
Chichewa	nya	Africa	Niger-Congo	Latin	110.98	30,322
Dawro	dwr	Africa	Afro-Asiatic	Latin	116.98	29,579
Dholuo	luo	Africa	Nilo-Saharan	Latin	76.02	30,429
Ewe	ewe	Africa	Niger-Congo	Latin	95.86	30,164
Gamo	gmv	Africa	Afro-Asiatic	Latin	107.07	30,200
Gofa	gof	Africa	Afro-Asiatic	Latin	86.78	30,388
Gujarati	guj	South Asia	Indo-European	Gujarati	83.09	30,447
Haitian Creole	hat	Caribbean	Creole	Latin	104.03	30,597
Hausa	hau	Africa	Afro-Asiatic	Latin	99.60	30,717
Hiligaynon	hil	Southeast Asia	Austronesian	Latin	106.43	29,260
Hindi	hin	South Asia	Indo-European	Devanagari	99.09	30,438
Igbo	ibo	Africa	Niger-Congo	Latin	94.46	30,011
Kannada	kan	South Asia	Dravidian	Kannada	104.86	30,495
Kikuyu	kik	Africa	Niger-Congo	Latin	87.08	30,722
Lingala	lin	Africa	Niger-Congo	Latin	128.49	28,790
Luganda	lug	Africa	Niger-Congo	Latin	101.75	30,440
Malayalam	mal	South Asia	Dravidian	Malayalam	86.12	30,190
Marathi	mar	South Asia	Indo-European	Devanagari	92.35	30,573
Ndebele	nde	Africa	Niger-Congo	Latin	100.76	30,161
Nepali	nep	South Asia	Indo-European	Devanagari	106.63	30,347
Oromo	orm	Africa	Afro-Asiatic	Latin	99.16	30,413
Punjabi	pan	South Asia	Indo-European	Gurmukhi	86.16	30,496
Shona	sna	Africa	Niger-Congo	Latin	74.66	30,685
Swahili	swh	Africa	Niger-Congo	Latin	96.38	30,634
Tamil	tam	South Asia	Dravidian	Tamil	92.93	30,516
Telugu	tel	South Asia	Dravidian	Telugu	93.18	30,059
Turkish	tur	Middle East	Turkic	Latin	63.50	29,747
Twi (Akuapem)	twi	Africa	Niger-Congo	Latin	71.20	30,270
Twi (Asante)	twi	Africa	Niger-Congo	Latin	78.87	30,565
Urdu	urd	South Asia	Indo-European	Arabic	88.26	30,634
Vietnamese	vie	Southeast Asia	Austroasiatic	Latin	71.62	30,451
Yoruba	yor	Africa	Niger-Congo	Latin	89.91	30,625
Total					3,469.01	1,121,956

Dataset Structure

Each language is a separate config with a single train split. One row = one verse:

Field	Type	Description
`audio`	audio	Verse-level mono waveform (sampling rate varies by language)
`text`	string	Verse transcript, original punctuation preserved
`testament`	string	`Old Testament` or `New Testament`
`book` / `chapter` / `verse`	string	Scripture reference
`duration_seconds`	float	Clip duration
`speaker_id`	string	Narrator label from diarization

Usage

python

from datasets import load_dataset

ds = load_dataset("multilingual-tts/open-bible", "Yoruba")

Limitations

Read, scripture-domain speech with a narrow vocabulary and formal register, models may not transfer well to conversational speech. Speaker labels and forced alignments are automatic and may contain errors.

License

All audio and transcripts come from Open Bible under CC BY-SA; derivatives must be shared under compatible terms with attribution retained. Provided as a speech-technology resource only.

Citation

Please cite the following if this was useful for your work:

@misc{guzmán2026openbiblettslargescalespeechresources,
      title={OpenBibleTTS: Large-Scale Speech Resources and TTS Models for Low-Resource Languages}, 
      author={David Guzmán and Luel Hagos Beyene and Jesujoba Oluwadara Alabi and Yejin Jeon and Dietrich Klakow and David Ifeoluwa Adelani},
      year={2026},
      eprint={2606.09553},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2606.09553}, 
}