Dec 02, 2020, 13:39 pm
Hey all, I know some individuals appreciate subtitles and while it would still be a bit of a pain, I think it's well within the realm of possibility creating RiffTrax-only subtitles or RiffTrax+movie subtitles by taking advantage of AI-assisted transcribing. This isn't a project I'm particularly interested in pursuing myself but, I wanted to inform others who may be interested. I've tested fireflies.ai and otter.ai which are websites which auto-transcribe audio files to text. Otter.ai is much better and has a reasonable price (Basically $13 for 6,000 minutes - and half that if you have a student email). They provide timing information and auto-identify speakers as well. Because it's AI, subsequent uploads should be more accurate after every upload and correction process. The paid plan it even has custom word support (so you could add "RiffTrax" and "Disembaudio" to your word list). I uploaded the RiffTrax audio file of Starship Troopers. I went through the first 10 minutes of the Starship Troopers and corrected it so you can see a before and after (which took about 10 minutes). With the "before", all I did was spend 60 seconds identifying the speakers:
Otter.ai-only Before: https://pastebin.com/ap5hizRe
After Human Editing: https://pastebin.com/MVD6Xqkd
Now, even after editing, that doesn't get it into a proper subtitle format but that shouldn't be too difficult once the transcription is done. Figuring out the ending time of the subtitle would require a little bit of work but some computer-guessing could probably be done, then manual touch-ups afterwards. Because the AI system also identifies different speakers, it would be trivial on top of everything else to create SSA subtitles which allow for each speaker (and the movie itself) to have a different subtitle text color for instance.
Anyway, if someone needs a project, here you go! I'd be happy to help with the 'getting it into the subtitle format' once the human-corrections and editing is done if someone needs help with that piece.
Otter.ai-only Before: https://pastebin.com/ap5hizRe
After Human Editing: https://pastebin.com/MVD6Xqkd
Now, even after editing, that doesn't get it into a proper subtitle format but that shouldn't be too difficult once the transcription is done. Figuring out the ending time of the subtitle would require a little bit of work but some computer-guessing could probably be done, then manual touch-ups afterwards. Because the AI system also identifies different speakers, it would be trivial on top of everything else to create SSA subtitles which allow for each speaker (and the movie itself) to have a different subtitle text color for instance.
Anyway, if someone needs a project, here you go! I'd be happy to help with the 'getting it into the subtitle format' once the human-corrections and editing is done if someone needs help with that piece.