rishi70612/nepali_asr

Dataset Card for Nepali Asr Dataset

Dataset Summary

This dataset consists of over 5 hours (300+ minutes) of English speech audio collected from YouTube. The dataset is designed for automatic speech recognition (ASR) and speaker identification tasks. It features both male and female speakers, with approximately 60% of the samples from male voices and the remaining 40% from female voices. The dataset contains 35 distinct speakers, each with their audio segmented into 30-second chunks.

Each audio sample is paired with a transcription and relevant metadata, making it suitable for training and evaluating ASR models as well as speaker identification systems.

---

Dataset Structure

• validation_dataset/
  • speaker_1/
    • speaker_1_chunk_001.wav
    • speaker_1_chunk_002.wav
    • ...
  • speaker_2/
    • speaker_2_chunk_001.wav
    • ...
  • ...
• validation_transcriptions.tsv
  (Columns: utterance_id, speaker_id, utterance_path, transcription, num_frames)

Example directory structure:

validation_dataset/
├── speaker_1/
│   ├── speaker_1_chunk_001.wav
│   ├── speaker_1_chunk_002.wav
│   └── ...
├── speaker_2/
│   ├── speaker_2_chunk_001.wav
│   └── ...
└── ...
validation_transcriptions.tsv

---

Data Collection

• Source: All audio was collected from YouTube.
• Speakers: 35 total (approx. 60% male, 40% female)
• Duration: 5+ hours (300+ minutes)
• Chunking: Each speaker's data is divided into 30-second audio chunks.

---

Data Fields

• utterance_id: Unique identifier for each audio chunk.
• speaker_id: Anonymized speaker label.
• utterance_path: Relative path to the audio file.
• transcription: Text transcription of the audio.
• num_frames: Number of frames in the audio chunk.

---

Intended Use

• Training and evaluation of automatic speech recognition (ASR) models.
• Speaker identification and diarization research.
• Audio processing and speech analysis tasks.

---

Example Usage

from datasets import load_dataset

dataset = load_dataset("/rishi70612/nepali_asr")
print(dataset["validation"][0])

---

Citation

Please cite this dataset as follows (update with your citation details):

@dataset{rishi70612/nepali_asr,
  title = {Nepali Asr},
  author = {Rishikesh Kumar Sharma},
  year = {2025},
  url = {https://huggingface.co/datasets/rishi70612/nepali_asr}
}

---

Additional Notes

• Ensure you comply with YouTube's terms of service and copyright policies when using or distributing this dataset.
• The dataset may contain inherent biases related to speaker selection or YouTube content.