AI Script Corrections
R
Robert McKay
The ask AI feature would be vastly more helpful if you could ask it to correct likely transcription, punctuation, capitalization and paragraph break errors in the transcript and then correct rather than overdub the transcript when you apply the edits. Overdubs get rid of video and create all sorts of issues. It would be much better to be able to correct the transcript en masse. This is something I have always done in an external app like I have copy pasted the transcript into Notion or ClickUp, asked the AI there to fix it, and then manually replaced the transcript of the file. It would be much better, if you have multiple files in a script, to be able to do all of this within Descript with one click.
Canny AI
Merged in a post:
AI to lessen transcription errors
Z
Ziloni Green
Please use AI more in the transcription to fix very common mistakes like two vs two, their vs there, incorrect capitalization, punctuation and more. The transcription inaccuracies are embarrassingly bad and very time consuming to fix.
Canny AI
Merged in a post:
Proofread script after editing
J
Julie
AI does a great job with editing the script/video for clarity and removing filler words, but it leaves the script a little messy (lowercase sentences, etc). I'm able to ask AI to proofread the script, but it doesn't allow me to replace the existing script with the updates. That would be an excellent follow-up feature to include after removing of filler words or clean up is finished.
J
Jesús López
I'm experiencing the same issue. I want to clarify that what Robert McKay said can be found in this recording: https://www.loom.com/share/bcc665c0c7f7454380c079f9d58fa36d?sid=4c47f520-1cc4-4f0a-ab3c-32922f9048a8. I need a transcription for later dubbing, and I plan to use a custom algorithm for it.
I believe that the Underlord should correct the errors Robert mentioned, as manually correcting them is currently the only option.
To provide more context, my current workflow is as follows:
- Shorten word gaps
- Remove retakes
- Export video file
- Use Whisper, an OpenAI model, to transcribe
- Correct any mistakes in the transcription, being careful about mismatches in timestamps
By the way, Whisper does a great job of accurately separating new lines.
I hope that Descript could work on this feature to further improve the product.
Canny AI
Merged in a post:
Improve Quality of Transcripts
M
Marvin Schneider
High quality transcripts is where it all starts in the reuse of content.
While Descript does a pretty good job at literally transcribing what is said, it does a pretty poor job at speaker identification, grammar, punctuation, sentence structure, etc. At the moment, I cannot use the raw transcript for closed caption creation. I need to export the transcript into a word document, and go through it with a fine tooth comb, and then re synchronize with the audio. (Kudos for this feature Descript!).
Please use AI to process the raw transcript in to a high quality transcript that can be read as a stand alone document at college level quality. Why not use the AI to transcription model that converts the spoken word (with all the attendant speaking foibles) into a clean, grammatically correct transcript? Even better, learn from all of the "before" and "after" transcripts from the user. If I am going to spend four hours offline editing a one hour podcast, I would sure like that effort to pay off in terms of getting a high quality transcript for the next podcast!