MooreSpeechCorpora collects publicly available Mooré speech, aligned with its text, for speech recognition and text-to-speech research. This version holds tales, riddles, proverbs and the Bible. Requests are reviewed by hand, usually within a few days. The recordings come from third-party publications: Mooré oral literature (mooreburkina.com and its ipsapps pages, published by Urs Niggli) and the Mooré Bible on bible.com (text © Bible Society of Burkina Faso). They are shared for non-commercial research only and remain the property of their owners. License: CC BY-NC 4.0 for our annotations only; the audio and texts stay the property of their owners.

MooreSpeechCorpora

Publicly available Mooré speech, aligned with its text, for speech recognition and text-to-speech.

En bref. MooreSpeechCorpora rassemble la parole en mooré disponible publiquement, découpée en phrases et alignée avec son texte, pour la reconnaissance et la synthèse vocales. Cette version contient des contes, devinettes et proverbes (7,7 heures) et la Bible en mooré de bible.com (74 heures). L'accès est sur demande, pour la recherche non commerciale. Une version plus propre existe : MooreSpeechCorporaV2.

MooreSpeechCorpora aims to collect all publicly available Mooré (ISO 639-3 mos) speech, cut into sentences and paired with its text, for speech recognition and text-to-speech research. This first version holds read tales, riddles and proverbs, which bring together the three datasets we first published separately, and the Mooré Bible from bible.com. Next versions add more sources.

The same work now extends to other languages of Burkina Faso (Dioula, Fulfuldé, Gourmantchéma and more) through FasoSpeechCorpora, the toolkit that builds these corpora.

For text-to-speech, prefer MooreSpeechCorporaV2, rebuilt from the same recordings with music trimmed and every transcript checked.

This is the GO AI Corp copy of anyantudre/MooreSpeechCorpora, kept identical.

Access is gated and requests are reviewed by hand. Anonymous or vague requests will not be approved.

Overview

default: 5,020 clips, 7.70 hours of Mooré, 16 kHz mono, with a denoised copy of each clip
2 speaker labels (see Limits)
The contes, devinettes and proverbes subsets keep the original datasets' clips and columns, including the French proverb readings, with a reader voice column added
bible: 57,410 clips, 74.28 hours of the Mooré Bible read aloud

Subset	Content	Splits	Clips	Hours
`default`	Mooré clips of all three categories, denoised copy, `category` column	`train`	5,020	7.70
`contes`	Tales	`vol2` to `vol5`	5,958	4.81
`devinettes`	Riddles (`train`) and proverbs `prv-v09` (`vol9`, see Limits)	`train`, `vol9`	879	2.14
`proverbes`	Proverbs, Mooré audio and text paired with the French reading	`vol1` to `vol11`	1,323 pairs	2.78 Mooré, 2.83 French
`bible`	Mooré Bible WẼNNAAM SEBRE 2014 from bible.com, 39 books, 929 chapters	`train`	57,410	74.28

default is not the sum of the subsets. It leaves out contes vol5 (badly cut: median CER 0.73), the French readings and proverbes vol9. It also leaves out the Bible for now: its owner and terms differ, and it will join default in a later version once it is denoised and labelled like the rest.

The jw.org recording of the Mooré Bible is not included: jw.org's terms of use do not allow its content to be posted online.

Quick start

from datasets import load_dataset

ds = load_dataset("goaicorp/GOAI-MooreSpeechCorpora", split="train")             # default
tales = load_dataset("goaicorp/GOAI-MooreSpeechCorpora", "contes", split="vol3")  # one subset, one volume
proverbs = load_dataset("goaicorp/GOAI-MooreSpeechCorpora", "proverbes", split="vol10")
bible = load_dataset("goaicorp/GOAI-MooreSpeechCorpora", "bible", split="train")

Log in first (hf auth login) with an account that has been granted access. The version first published here, with train, test and full splits, stays available with revision="v1.0". Its test split shares 172 clips with train, so do not use it for evaluation.

Columns

default

Column	Type	Description
`audio`	Audio (16 kHz)	The clip
`denoised_audio`	Audio (16 kHz)	The same clip with background music reduced (Resemble Enhance)
`text`	string	Mooré transcript
`duration`	float	Length in seconds
`voice`	string	Reader's voice: `male` or `female`
`speaker_id`	int	1 or 2
`category`	string	`conte` (tale), `devinette` (riddle) or `proverbe` (proverb)

contes and devinettes: audio, transcription, duration, chapter (book, chapter and page of the source), voice (reader: male or female), plus speaker_id in devinettes.

proverbes: moore_audio, moore_text, french_audio, french_text, duration_moore, duration_french, chapter, moore_voice (Mooré reader: male or female). In vol1 to vol8 each proverb is read twice in Mooré, so two rows share one French clip.

bible: audio, transcription, duration, chapter (book, chapter and version, e.g. GEN_1_MPBU). The reading is dramatised with several voices, so there is no voice column yet.

Sources and credits

Tales, riddles and proverbs. All recordings were published by Urs Niggli on mooreburkina.com and its offline pages on media.ipsapps.org/mos/ora/, and in the "Sagesse Africaine" and "Burkina Langues" apps.

Part	Source	Authors and readers	Clips
contes `vol2`	Contes mooré vol. 2, 18 tales (`co2`)	Written by Jacques Siballo	550
contes `vol3`, `vol4`, `vol5`	Tale books from the same site; `vol3` is probably Contes mooré vol. 1 (`co1`)	`co1`: Jacques Siballo, read by Ruth Ouedraogo (2021)	5,408
devinettes `train`	Probably Devinettes en mooré (`devin`, 291 riddles)	Not credited on the site	611
proverbes `vol1` to `vol8`	Proverbes en mooré vol. 1 to 8 (`p1` to `p8`)	French translation by Timothée Kaboré	644
proverbes `vol9` to `vol11`, devinettes `vol9`	Yel-bũna vol. 1 to 3 (`prv-v09` to `prv-v11`)	Collection attributed to Alain Joseph Sissao (from the audio file names)	947

We thank Urs Niggli, the authors and the readers for making Mooré oral literature available online. Alignment used Meta's MMS models; the denoised audio was made with Resemble Enhance.

Bible. WẼNNAAM SEBRE 2014 (version 3058, MPBU) on bible.com. Text: Moore Bible © Bible Society of Burkina Faso, 2014. The audio's rights holder is not stated on bible.com. bible.com's terms allow personal, non-commercial use only.

How it was built

The original pipeline (fairseq MMS forced alignment, then Resemble Enhance) now lives in FasoSpeechCorpora as the fairseq backend and tutorials/noise_reduce.ipynb.

Download each book's audio and text, convert to 16 kHz mono.
Cut the audio into sentences by forced alignment with the text.
For default: keep the Mooré side, denoise each clip, label speakers.
For bible: each chapter's audio and text from bible.com, aligned with the fairseq MMS aligner (the bible.com source in FasoSpeechCorpora).
November 2026 reorganization: the three datasets were moved in as subsets, with a reader voice column added to contes and proverbes (one reader per book; V1 labels and V2 voice embeddings agree). In default, 515 proverb clips held twice (vol10 and vol11, identical audio) were removed and a category column was added. The bible.com part of anyantudre/moore-speech-bible was copied in as the bible subset, unchanged.

Limits

Music: many clips keep a few seconds of music at the start or end (riddles start with 3 to 9 s of jingle). V2 trims them.
Alignment: contes vol5 is misaligned (clips often hold the end of the previous sentence; 400 clips are under 0.5 s). Kept in the contes subset for reference, left out of default.
Transcripts: some have split words, stray hyphens, leading numbers or a proverb written twice.
Speakers: speaker_id 1 groups two different men (the tales vol2 reader and the proverbs vol1 to vol8 reader); speaker_id 2 is one woman. The labels were not checked by ear.
devinettes vol9 holds proverbs from Yel-bũna vol. 1, not riddles. It is labelled proverbe in default's category column.
bible: not yet trimmed, checked or denoised like V2. 406 clips are longer than 20 s (up to 86 s), 38 are shorter than 0.5 s and 1,970 texts appear more than once. Religious language only.
Content: this version has read speech only (oral literature and the Bible), no conversational speech.

Access

Fill in the form on this page. Requests are approved for people and institutions that say who they are, describe a research or academic use, and accept the non-commercial terms. Questions: Wendyellé A. Alban NYANTUDRE (nyantudre@gmail.com).

License

Audio and source texts: property of their authors, readers and publishers (see Sources and credits); the Bible text belongs to the Bible Society of Burkina Faso. No license is granted on them; they are shared for non-commercial research and their owners can ask for removal.
Our work (segmentation, alignment, corrected transcripts, metadata): CC BY-NC 4.0.

For other uses, contact us. By accessing this dataset you agree to these terms.

Citation

@misc{nyantudre2025moorespeechcorpora,
  title        = {MooreSpeechCorpora: a Mooré speech corpus for TTS and ASR},
  author       = {Nyantudre, Wendyellé Abubakrh Alban and Bandaogo, Aristide},
  year         = {2025},
  howpublished = {Hugging Face Datasets (gated access)},
  url          = {https://huggingface.co/datasets/anyantudre/MooreSpeechCorpora}
}

Contacts

Wendyellé A. Alban NYANTUDRE, curator: nyantudre@gmail.com, LinkedIn
Aristide BANDAOGO: aristide@goaicorporation.org

Downloads last month: 135

Total file size:

7.7 GB

Collection including goaicorp/GOAI-MooreSpeechCorpora

Mooré Speech Datasets 🇧🇫💛

Collection

Our Mooré language datasets. • 2 items • Updated 1 day ago • 2