onlysainaa/common-voice-mn-24

🇲🇳 Common Voice Mongolian 24.0 Dataset

This repository hosts the latest release (v24.0) of the Mozilla Common Voice Scripted Speech dataset for Mongolian (mn). This dataset is a vital resource for training robust Automatic Speech Recognition (ASR) and Text-to-Speech (TTS) systems for the Mongolian language.

---

📊 Dataset Statistics

Metric	Value
Total Clips	96,308
Total Duration	140.56 Hours
Validated Duration	49.19 Hours
Total Speakers	606
Format	MP3 (contained in .tar.gz)

This version includes a significant increase in validated hours compared to previous releases, improving its reliability for supervised learning.

---

🚀 Getting Started

Method 1: Using huggingface_hub (Recommended for Raw Archive)

You can download the specific tarball programmatically. This method is best if you need the original directory structure.

from huggingface_hub import hf_hub_download
import os

repo_id = "onlysainaa/common-voice-scripted-speech-24.0-mongolian"
filename = "cv-corpus-24.0-2025-12-05-mn.tar.gz"

# Download the file
download_path = hf_hub_download(
    repo_id=repo_id,
    filename=filename,
    repo_type="dataset",
    local_dir="./data"
)

print(f"File downloaded to: {download_path}")
Method 2: Command Line (Git LFS)
Bash
git lfs install
git clone [https://huggingface.co/datasets/onlysainaa/common-voice-scripted-speech-24.0-mongolian](https://huggingface.co/datasets/onlysainaa/common-voice-scripted-speech-24.0-mongolian)
🛠️ Extraction & Usage
To extract the audio clips and metadata files:

Bash
tar -xf ./data/cv-corpus-24.0-2025-12-05-mn.tar.gz
Common use cases:

ASR: Fine-tuning Whisper or Wav2Vec2 models for Mongolian.

TTS: Developing natural-sounding Mongolian synthetic voices.

Linguistics: Analyzing dialectal variations and phonetic patterns.

📜 Citation If you use this dataset in your research or project, please cite it as follows:

Code snippet @misc{commonvoice_mn24, title = {Mozilla Common Voice 24.0 - Mongolian}, author = {Mozilla Foundation}, year = {2025}, publisher = {Hugging Face}, journal = {Hugging Face Datasets}, howpublished = {\url{https://huggingface.co/datasets/onlysainaa/common-voice-scripted-speech-24.0-mongolian}} }

🙏 Acknowledgements

Special thanks to the Mozilla Common Voice contributors and the Mongolian community for making this open-source speech data available.

A massive thank you to Jargalsaikhan Orgil [https://mn.linkedin.com/in/orgil-jargalsaikhan-56818236] from MUST-SICT, who has been the biggest motivation and driving force behind the collection of the Mongolian speech dataset in Common Voice. His dedication is one of the biggest reason this dataset exists today.

---