Inspiration

Watching recorded lectures is one of the most dreary parts of uni life. After a long day working on assignments, part time work, or other ways to spend the valuable daylight hours, maintaining the focus to watch (and actually digest) 2 hours of lectures is a big ask.

What it does

Lec2Note turns lectures into frame-synchronised transcripts to support your learning experience. Through these transcripts, students can:

  • Search through lectures and make sure nothing's missed
  • Skip silences and errata, and process lectures in idle time
  • Match narration with slides without the manual effort

How we built it

Lec2Note uses Google's Speech to text API and OpenCV.

  • Google speech to text is a cost effective way of getting high quality transcriptions from audio.
  • OpenCV uses local processing power to identify key turning points in the video stream.
  • ffmpeg and ffprobe were used to help with file type conversions. All of these tools are reasonably cross platform! ## Challenges we ran into
  • Creating a video pipeline that was reliable enough to split chunks of video without over-splitting busy periods or missing key transitions was quite difficult.
  • File format conversions are messy. Even things like # of audio channels tripped us up. ## Accomplishments that we're proud of Getting a robust video and audio processing pipeline up!

What we learned

Using opencv skills to create an actual real product was fun! We also learnt a few things about nodejs express and how to use a file system as a database in a pinch XD

What's next for lec2note

Popping it on a website so other people can use it would be nice :D We could also use wav2vec in the backend so that we can operate for free (sorry google)

Misc

You need to add your google credentials file as a google-credentials.json in the lec2note_main directory to use this.

Built With

Share this project:

Updates