BibleTTS Hausa - Aligned Bible Speech

Source

The Hausa subset of BibleTTS is derived from the Bible recordings released by Open.Bible and published by Biblica. CLEAR Global produced the first BibleTTS Hausa release by force-aligning those recordings to the Hausa Bible text, but its segmentation contained many alignment errors. This release force-aligns and re-segments the same recordings again.

Data Structure

• clips/ — 43,405 WAV files, 24 kHz mono PCM, named __.wav.
• metadata.tsv — one row per clip with the columns path, split, text, duration, score, speaker.

The corpus has two speakers, split by book rather than interleaved: speaker one reads 28 books (22,487 clips, 56.40 h) and speaker two reads 38 books (20,918 clips, 45.06 h). No clip mixes speakers.

score is the forced aligner's confidence for the clip (an average per-token log-probability; negative, higher is better, length-normalised). The train, dev and test splits are the split assignment shipped with this release, held out at book level: dev draws entirely from the book of Ezra and test entirely from the book of Colossians. The other split holds the 2,145 clips that appear in the corpus's full manifest but in none of the three splits; all have score <= -55 (median -73.6) and are not intended for TTS training as-is. Every clip is published in exactly one split.

Related resources

This corpus was used to train the Hausa text-to-speech models in the Hausa TTS models collection (VITS, XTTS, YourTTS).

Those TTS models were then used to synthesise additional Hausa speech, released as the Hausa Synthetic ASR Dataset (XTTS) and Hausa Synthetic ASR Dataset (YourTTS), and that synthetic audio was used to train some of the models in the Hausa W2v-BERT 2.0 and Hausa MMS-1B collections. This corpus was therefore never used to train ASR models directly, but those models descend from it: the chain is this corpus → TTS models → synthetic speech → ASR models.

Citation

Meyer, J., Adelani, D., Casanova, E., Öktem, A., Whitenack, D., Weber, J., KABONGO KABENAMUALU, S., Salesky, E., Orife, I., Leong, C., Ogayo, P., Chinenye Emezue, C., Mukiibi, J., Osei, S., AGBOLO, A., Akinode, V., Opoku, B., Samuel, O., Alabi, J., Muhammad, S.H. (2022) BibleTTS: a large, high-fidelity, multilingual, and uniquely African speech corpus. Proc. Interspeech 2022, 2383-2387, doi: 10.21437/Interspeech.2022-10850

@inproceedings{meyer22c_interspeech,
  title     = {{BibleTTS: a large, high-fidelity, multilingual, and uniquely African speech corpus}},
  author    = {Josh Meyer and David Adelani and Edresson Casanova and Alp Öktem and Daniel Whitenack and Julian Weber and Salomon {KABONGO KABENAMUALU} and Elizabeth Salesky and Iroro Orife and Colin Leong and Perez Ogayo and Chris {Chinenye Emezue} and Jonathan Mukiibi and Salomey Osei and Apelete AGBOLO and Victor Akinode and Bernard Opoku and Olanrewaju Samuel and Jesujoba Alabi and Shamsuddeen Hassan Muhammad},
  year      = {2022},
  booktitle = {{Interspeech 2022}},
  pages     = {2383--2387},
  doi       = {10.21437/Interspeech.2022-10850},
  issn      = {2958-1796},
}

DeRenzi, B., Dixon, A., Farhi, M.A., Resch, C. (2025) Synthetic Voice Data for Automatic Speech Recognition in African Languages. In Proceedings of the First Workshop on Advancing NLP for Low-Resource Languages, pages 152–186, Varna, Bulgaria. INCOMA Ltd.

@inproceedings{derenzi-etal-2025-synthetic,
  title     = "Synthetic Voice Data for Automatic Speech Recognition in {A}frican Languages",
  author    = "DeRenzi, Brian and Dixon, Anna and Farhi, Mohamed Aymane and Resch, Christian",
  booktitle = "Proceedings of the First Workshop on Advancing {NLP} for Low-Resource Languages",
  month     = sep,
  year      = "2025",
  address   = "Varna, Bulgaria",
  publisher = "INCOMA Ltd., Shoumen, Bulgaria",
  url       = "https://aclanthology.org/2025.lowresnlp-1.16/",
  pages     = "152--186"
}

Disclaimer

The recordings are derived from the Open.Bible project by Biblica and are redistributed here as a derivative work. Force-alignment, segmentation and packaging were carried out by CLEAR Global, continuing the Masakhane BibleTTS collaboration. Two-speaker Bible-domain data: not representative of everyday spoken Hausa. Provided as-is for research and development.

Please check README.md for more information.