The timestamps are doing nothing
An SRT file spends most of its bytes on timing: an index number, two timestamps and a blank line for every caption. That is what a player needs and what a reader does not. If you are going to read the subtitles, search them, quote them, count the words or paste them into another tool, all of that scaffolding is noise.
Removing it is not the same as summarising — nothing is dropped except the timing. The words stay in the order they were spoken.
What the output looks like
One caption per line, in order, separated by newlines. A caption that was split across two lines in the original becomes a single line, because that split was a layout decision to keep the caption on screen, not a paragraph break.
Conversely, each cue is its own line even if the original text was one continuous sentence — subtitle cues are cut to fit the screen, not to match sentence boundaries. If you need proper paragraphs, join the lines back together first.
What you lose
The timings, permanently — they are not recoverable from the text output. So the workflow matters: keep the original SRT, convert a copy.
If what you actually want is subtitles that a player can read, do not convert to TXT at all. And if the source text is what you have and you need timings rebuilt from scratch, that is the opposite job — see text to SRT.
What this is usually for
- Reading a transcript of a video without scrubbing through it.
- Finding a quote — search the text, and if you need the exact moment, go back to the SRT.
- Word counts and any kind of text analysis where punctuation noise like
00:00:01,000would be counted as tokens. - Feeding another tool that takes plain text — a summariser, a translator, a CMS field.
Source formats other than SRT
WebVTT, SBV and SAMI all convert to the same plain text here, because they all carry the same underlying thing: text plus timings. Which format it came from does not change the result.