Massive YouTube Educational Transcriptions with 35,890 Audio Hours

A large-scale collection of 59,355 educational YouTube video transcriptions generated using the distil-whisper/distil-large-v3.5 model. The dataset contains approximately 1.54 billion characters of transcribed text and was created by thepowerfuldeez, with a last recorded update on 2026-02-21.

Use Cases

• Train or fine-tune speech recognition models based on the large volume of transcribed educational audio.
• Analyze educational discourse and content structure based on the full transcript text.
• Build search or recommendation systems for educational videos based on the transcribed content.
• Conduct linguistic or topic analysis of online educational material based on the video titles and transcripts.

Strengths

• Large scale with 59,355 videos and 35,890 hours of transcribed audio.
• Transcribed using a specific, modern model (distil-whisper/distil-large-v3.5).
• Processing was performed efficiently, at 165-185x realtime speed on specified hardware.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect source and topical bias inherent to the YouTube platform and the creator's selection criteria.

Provenance

Source

YouTube

Collection Method

Transcribed using faster -whisper (CTranslate2) with the distil-large-v3.5 model.

Time Range

null

Freshness

Last updated 2026-02-21 06:01:19; freshness should be verified.

Geography

null