README.md

metadata

language:
  - mos
license: cc-by-4.0
size_categories:
  - 10K<n<100K
task_categories:
  - text-to-speech
  - automatic-speech-recognition
pretty_name: Moore Speech Bible
dataset_info:
  - config_name: moore
    features:
      - name: audio
        dtype: audio
      - name: transcription
        dtype: string
      - name: duration
        dtype: float32
      - name: chapter
        dtype: string
    splits:
      - name: train
        num_bytes: 5457469434.45
        num_examples: 57410
      - name: validation
        num_bytes: 20117104.9
        num_examples: 194
    download_size: 4321991415
    dataset_size: 5477586539.35
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*

Dataset Card for Moore Speech Bible (Private Internal)

Access: This dataset is hosted privately in the GO AI Corporation Hugging Face organization. Access is limited to approved collaborators under YouVersion’s non-commercial terms.

Dataset Description

• Languages: Mooré (ISO 639-3: mos)
• Total Audio: ≈ 80 hours
• Repository: Moore Speech Corpora Toolkit
• Supported Tasks: Text-to-Speech (TTS), Automatic Speech Recognition (ASR)

Summary

The Moore Speech Bible dataset contains sentence-level aligned audio and transcripts from the Mooré Bible (“WẼNNAAM SEBRE 2014”) on bible.com:

• Audio: 16 kHz mono WAV
• Transcripts: Preserve Mooré diacritics
• Alignment: Fairseq MMS forced-aligner, with Uroman for romanization

All code is available in the Moore Speech Corpora Toolkit. ⭐ if it’s helpful!

---

Dataset Structure

Data Instances

{
  "audio": {
    "path": "data/train/GEN_1_MPBU_segment001.wav",
    "array": [/* waveform samples */],
    "sampling_rate": 16000
  },
  "transcription": "Wakat-kãnga, mam na n tigm-y lame.",
  "duration": 3.16,
  "chapter": "GEN_1_MPBU"
}

Fields

• audio: Path & decoded waveform at 16 kHz
• transcription: Aligned Mooré text
• duration: Segment length in seconds
• chapter: Original chapter code

Splits

Split	Examples	Approx. Duration
train	57 410	~76 hrs
validation	194	~0.5 hrs

---

Dataset Creation

1. Crawling

   • crawlers/bible_crawler.py (Scrapy) fetched MP3s and transcripts from bible.com.

2. Preprocessing

   • Resampled to 16 kHz WAV (preprocessing/resample.sh).
   • Cleaned/normalized text (preprocessing/text_normalization.py).

3. Alignment

   • Fairseq MMS forced-align with Uroman (forced_alignement/align_and_segment.sh).

4. Packaging

   • Converted to HF Dataset via datasets/prepare_hf_dataset.py.

Note: Usage follows YouVersion’s non-commercial, internal-use license.

---

Contacts & Citation

Curator:

• Wendyellé Abubakrh Alban Nyantudre (anyantudre)

Citation:

@misc{anyantudre2025moorespeechbible,
  title        = {Moore Speech Bible: A Curated Aligned Audio/Text Dataset for Mooré TTS/ASR},
  author       = {Nyantudre, Wendyellé Abubakrh Alban and Bandaogo, Aristide},
  year         = {2025},
  howpublished = {Gated dataset, GO AI Corporation Hugging Face org}
}

Considerations:

• Legal: Internal, non-commercial research only. Redistribution outside GO AI is prohibited without permission.
• Bias: Domain-specific (Bible readings).
• Speakers: Single-narrator; limited variability.