an very high portion of the videos we all make includes product names that might be words mashed together or all new words entirely.
Being able to help the transcription by teaching it a few names that will come up in the project should help the AI drastically. If I'm making a video for The Baconator at Wendy's and it's always transcribing interviews as "Bacon ate her at when these" it's a much less useful tool. Giving it the headstart of those words would give the model something to work around.
I suggest implemetation could be a text window where you can type some common terms, tab-delimited.