Get Eric Jadi’s stories in your inbox
Join Medium for free to get updates from this writer.
The project relies on Azure Cognitive Services for the actual speech analysis and makes use of FFMPEG for video manipulation and Youtube-DL to download videos.
This prototype has been packaged as a Docker container, so anyone with a Linux Docker environment and access to Azure Cognitive Services can run this themselves.
On a very high level, the process can be broken down into the following steps:
1. Download the YouTube video in mp4 format.
2. Extract the audio and cut this into 10 second segments.
3. Feed the 10 second audio segments in parallel into Azure Speech Services for per-word timestamps.
4. Mark all subsequent instances of the words “You” and “Know” and generate a full timestamped list of which cuts to make.
5. Finally, cut out all the marked segments and put it back together into one file.
For the technically inclined: I’ve open sourced the project on GitHub where you can find a more in-depth technical explanation of how the prototype works and how to use it.
So, you know, have fun!