Let me preface this with: I’m a huge fan of pretty much all things Rooster Teeth.
A while back I heard it through the grapevine, that the existing Rooster Teeth Podcast database had gone out of commission.
Since I had my daughter, I had been needing a good ol’ fashioned side project, so this seemed like a slam dunk.
During my research into how I was going to make the site, I found quite a few Rooster Teeth Podcast databases, so to set mine apart from the others, I decided that my version of the database would also search transcripts of the episodes.
This of course meant that I had to transcribe over 400 individual episodes. A seemingly insurmountable task.
I decided to use Google Cloud Speech and Amazon Transcribe as the driving force of the transcriptions, but as with everything extremely complicated (like transcribing hundreds of hours of speech), the work done by a computer had to be proofed by a human. So not only did I have to build a transcription engine, but I had to build an interface for users to proofread the generated text. And because there is a human element to the equation, there has to be other humans to make sure the first humans don’t deliberately mess things up.
I believe that speech recognition and human proofreading is the most effective way of transcribing, and that is still my belief after having finished this project.