Skip to content

Latest commit

 

History

History
46 lines (27 loc) · 8.93 KB

README.md

File metadata and controls

46 lines (27 loc) · 8.93 KB

ScribeSalad

In absence of searchable transcripts, many interesting YouTube videos, podcasts, lectures and talks are hard to explore, quote and summarize. ScribeSalad is a multi-lingual open data project regrouping over 940k YouTube video transcripts discussing social and political issues, psychology, history and scientific topics ranging from biology, mathematics to artificial intelligence : TedX, Yale courses, MIT lectures, National Geographic, The Joe Rogan Experience, Big Think, IQ squared, Jordan B. Peterson talks, Tim Ferris, Jocko Podcast and more. This project is a first step towards making great content more available and inspiring speakers, storytellers, interviewers and scientists better heard.

Available transcripts (in english)

Other languages

Arabic (ar), French (fr), German (de), Spanish (es), Russian (ru), Turkish (tr), Portuguese (pt), Italian (it), Japanese (ja), Korean (ko)

Transcription quality

Some of the transcriptions originate from YouTube (subtitles uploaded by the video's owner) while the rest are generated automatically using a high-accuracy large-vocabulary continuous speech recognition system (~90% of accuracy in clean conditions : no background noise, no heavy accents and good quality audio).

Filenames and formats

The transcripts identified using the corresponding YouTube videos IDs and each one is available in three formats : text, vtt (Text Tracks Format) and srt (SubRip Subtitle Format).

To open the original video, replace "ID" in https://www.youtube.com/watch?v=ID by the transcript filename.

Terms of use

This is an open data project, feel free to fork this repository, download, share and use any of the transcripts.

TODO

  • Cleaning-up transcripts : removing fillers (hum, ah, etc) and repetitions.
  • Re-aligning transcripts : re-aligning transcripts and fixing overlapping timecodes.
  • Topic modeling : automatically discovering the abstract "topics" that occur in a each transcript.
  • Speaker identification : who spoken when ? and for how long ?
  • Creating a search engine : exploring subjects by speaker, topic, channel, etc.
  • Multiligual transcripts : Translating all transcripts to other languages.
  • More channels & more videos.