---
license: mit
pretty_name: yt transcripts on sn13 beginning
size_categories:
- 10K<n<100K
---

<center>
    <img src="https://huggingface.co/datasets/macrocosm-os/images/resolve/main/bittensor.png" alt="Data-universe: The finest collection of social media data the web has to offer">
</center>

<center>
    <img src="https://huggingface.co/datasets/macrocosm-os/images/resolve/main/macrocosmos-black.png" alt="Data-universe: The finest collection of social media data the web has to offer">
</center>


# 🎙️ YouTube Transcripts Dataset

> **Created by [SN13](https://www.macrocosmos.ai/sn13)(, powered by [Bittensor](https://www.bittensor.com/)**

This dataset provides high-quality transcripts from **26,241 public YouTube videos**, meticulously collected and formatted for research and development in **natural language processing (NLP)**, **summarization**, and **language modeling**.

---

## 📦 Dataset Summary

- **Total Videos**: 26,241
- **Total Hours**: ~6,886.84
- **Language Coverage**: Multilingual with translation to English(details below)
- **Storage Format**: Parquet
- **File Splits**: 3 files, each ≤10,000 rows
- **Created By**: [SN13](https://www.macrocosmos.ai/sn13)
- **Powered By**: [Bittensor](https://www.bittensor.com/)

---

## 🧱 Schema

Each record represents one YouTube video with metadata and its full transcript:

| Column               | Type     | Description |
|----------------------|----------|-------------|
| `video_id`           | `string` | YouTube video ID |
| `title`              | `string` | Video title |
| `channel_id`         | `string` | Channel's unique ID |
| `channel_name`       | `string` | Display name of the channel |
| `url`                | `string` | Direct link to the video |
| `transcript`         | `list`   | List of segments with start/end time and text |
| `language`           | `string` | Language code (e.g., `en`, `fr`, etc.) |
| `duration_seconds`   | `int64`  | Duration of the video in seconds |
| `upload_date.$date`  | `string` | Video upload date in ISO format |

---

## 💬 Transcript Format

Each transcript is a list of time-aligned text segments:

```json
[
  { "start": "0.760", "end": "4.600", "text": "Welcome to the channel." },
  { "start": "4.600", "end": "8.320", "text": "Let's dive into the details..." }
]