sourabsb/kumaoni-dataset

🗣️ English–Kumaoni Multimodal Dataset (Romanized + Devanagari + Audio, 9538 Entries)

This dataset contains 2112 sentence pairs consisting of:

• English text
• Kumaoni translation in Roman (Latin) script
• Kumaoni translation in Devanagari (Unicode)
• Corresponding audio file paths

It is designed to support both text-based and speech-based NLP tasks for the Kumaoni language, a low-resource Indo-Aryan language spoken in the Kumaon region of Uttarakhand, India.

---

📂 Dataset Format

• Format: JSONL (newline-delimited JSON)
• Encoding: UTF-8
• Total entries: 2112
• Fields:
  • english: Sentence in English
  • kumaoni_roman: Kumaoni sentence in Latin script
  • kumaoni: Kumaoni sentence in Devanagari script
  • audio_path: Path to corresponding audio file

📄 Example:

{
  "english": "give me water.",
  "kumaoni_roman": "mank paani diyo.",
  "kumaoni": "मंक पानी दियो।",
  "audio_path": "audio/0001.wav"
}

---

🏔️ About Kumaoni

Kumaoni (कुमाउनी) is an underrepresented regional language spoken in the Kumaon division of Uttarakhand, India. Despite over a million speakers, it lacks digital and NLP resources.

This dataset helps bridge that gap by providing:

• Clean bilingual sentence pairs
• Dual script support (Roman + Devanagari)
• Audio for pronunciation and speech modeling

---

🔍 Use Cases

• Machine Translation (English ↔ Kumaoni)
• Fine-tuning low-resource translation models (e.g., mBART, NLLB, ByT5)
• Speech-to-text (ASR) and text-to-speech (TTS) tasks
• Building Kumaoni chatbots and voice assistants
• RAG systems with multilingual grounding
• Multimodal language modeling for underrepresented languages

---

🔈 Audio Details

• Audio files are clear, human-recorded female voice samples
• Sampling rate: 16 kHz (recommended for most ASR/TTS pipelines)
• Audio filenames match the audio_path field

---

📜 License

CC BY 4.0
You may use, share, and modify this dataset with proper attribution to the original author.

---

🙏 Acknowledgements

Created and maintained by Sourab Singh Bora
This dataset focuses on clean Roman-Devanagari pairing, natural speech recordings, and Unicode accuracy.

---

💡 Citation

@dataset{english_kumaoni_multimodal_2025,
  author       = {Sourab Singh Bora},
  title        = {kumaoni-dataset},
  year         = 2025,
  url          = {https://huggingface.co/datasets/sourabsb/kumaoni-dataset},
  note         = {Parallel corpus with audio and dual-script support for low-resource translation and speech tasks}
}