goaicorp/new-dioula-speech-clean

Dioula Speech Corpus: A Parallel Audio-Text Dataset for Dioula and French

The Dioula Speech Corpus is a bilingual audio-text corpus designed for research and academic purposes in low-resource speech and language processing. It is intended primarily to support the development of Text-to-Speech (TTS) and Automatic Speech Recognition (ASR) models for the Dioula language.

⚠️ Access is gated. To request access, please read the policy below.
🛑 TLDR: For safety and traceability reasons, anonymous or vague requests will not be approved.

---

Dataset Structure

Dataset Overview

Property	Value
Splits	train
Total examples	~10,929
Languages	Dioula (dyu), French (fr)

Data Fields

Each record contains:

• audio — The audio file containing the spoken Dioula utterance.
• dyu — The text transcription in Dioula.
• fr — The corresponding French translation.

Example

{
  "dyu": "A bi ji min na",
  "fr": "Il boit de l'eau.",
  "audio": {
    "path": "common_voice_dyu_18897661.wav",
    "array": [...],
    "sampling_rate": 48000
  }
}

---

Use Cases

• 🔬 Research on low-resource speech technologies
• 🗣️ TTS and ASR system development for Dioula
• 🌍 Cross-lingual translation modeling (Dioula ↔ French)
• 📚 Digital preservation of West African languages

❌ Commercial use is not permitted without explicit authorization.

---

Access Policy

The dataset is under gated access to ensure ethical use and data integrity.

To request access, contact Wendpanga Aristide Bandaogo — aristide@goaicorporation.org

Access is granted only to individuals or institutions who:

1. Clearly identify themselves
2. Describe their intended academic or research use
3. Agree to the non-commercial usage restriction

🔐 Anonymous or vague requests will not be approved.

---

Contributing

We welcome collaboration from anyone passionate about under-resourced African languages in AI!

• 💡 Suggest dataset improvements or extensions
• 📝 Improve documentation or metadata
• ✅ Help with validation, alignment, or text normalization

Open an issue or contact us directly. Together, let's help African languages shine in AI! 💛

---

License

Licensed under CC BY-NC 4.0 — attribution required, strictly non-commercial.

For other usage scenarios, please contact us directly.

---

Citation

@misc{bandaogo2026dioulaspeech,
  title        = {Dioula Speech Corpus: A Parallel Audio-Text Dataset for Dioula and French},
  author       = {Bandaogo, Wendpanga Aristide},
  year         = {2026},
  howpublished = {Hugging Face Datasets (gated)}
}

---

Contact

Curator: Wendpanga Aristide BANDAOGO
📧 aristide@goaicorporation.org
🏢 GO AI CORPORATION