[Question] How to efficiently find and fix abnormal timings (0ms / extremely short lines) in Aegisub?

[zoqapopita93]

Hi everyone,

I'm currently working on an AI-translated SRT file (about 2000 lines). While the text quality is decent, the timing suffers from typical AI-generated flaws that severely impact the viewing experience. Specifically, I'm dealing with the following issues:

1. "Ghost" lines (0ms duration): Lines where the start and end times are exactly the same (e.g., 00:15:46,659 --> 00:15:46,659).
2. Extremely short "flashing" lines: A massive amount of brief interjections (like "Uh", "Oh", "Okay") lasting only 20ms to 100ms, making them impossible to read.
3. Screen flickering: Dialogue is chopped into too many tiny fragments, causing the subtitles to flash rapidly on screen during fast-paced conversations.

My current struggle:
I know I can sort the grid by "Duration" to manually fish out and delete the 0ms lines. However, manually locating, extending, or joining (merging) hundreds of lines under 100ms within a 2000-line file is incredibly tedious. There doesn't seem to be a straightforward "Error Checker" panel for this in Aegisub.

My questions:

1. Is there a built-in feature or a trick (perhaps using Regex in the "Select Lines" tool) to automatically select all lines with a duration below a specific threshold (e.g., Duration < 300ms)?
2. Are there any recommended Automation scripts (Lua macros) or established workflows in the community for cleaning up heavily fragmented AI timings? I'm hoping for something that can "batch merge adjacent short lines" or "automatically enforce a minimum duration."

Any advice or pointers to existing scripts would be greatly appreciated. Thanks in advance!

[arch1t3cht]

I'm not aware of any existing script for your use case but I can answer your first question: You can use the LuaInterpret script (available on DependencyControl) to write short snippets to select lines based on their duration. For your example, a simple snippet like if line.duration <= 300 then select() end (you'll need to select all lines using e.g. Ctrl+A before running it) would work.

[SUCM]

UI has a CPS column that can turn red for each subtitle line where the number of characters for the time duration is way too much. you should be able to adjust threshold in the setting