You finish a 90-minute narrated video. The script is done, the AI voiceover is exported, and the only thing left before upload is captions. So you search for how to turn a voiceover into an SRT file, and the results all say the same thing: run the audio through speech recognition, then fix what it got wrong.
That advice is built for people who have a recording and nothing else. If you wrote the script and the voice read it, you are solving a different, easier problem, and treating it like transcription costs you time and accuracy for no reason.
You Do Not Need to Transcribe a Script You Already Wrote
Speech recognition does two jobs at once: it works out which words were spoken, and it works out when. The first job is the hard one and the one that goes wrong. Names, places, numbers and invented terms are where recognizers guess. A name like "Hatshepsut" can come back as three unrelated words, and a brand or character name you coined will be spelled however the model hears it.
When you already have the script, the first job is done. Every word, every spelling choice, every comma is already correct. All that is missing is the timing: when each line starts and stops.
The cost of ignoring this adds up on long videos. At a typical narration pace of about 150 words a minute, a 90-minute video is roughly 13,500 words. If even 1 word in 100 comes back wrong, that is around 135 corrections you have to find by watching or reading the whole file. With your own script as the text, there are none to find, because the text never changed. We covered the broader question of what to look for in captioning software in our guide to AI subtitle generators for long-form YouTube.
Notice also that a voice reading your script may not say what is on the page. If a name is mispronounced, the captions should still show the correct spelling, which is another reason to keep the text yours. The pronunciation side is covered in fixing AI voice mistakes with names and numbers.
What an SRT File Actually Contains
An SRT is the simplest caption format there is: a plain text file made of numbered blocks. Each block has a start time, an end time and the words to show.
1
00:00:00,420 --> 00:00:03,110
Welcome back. Tonight we are going to the bottom of the sea.
2
00:00:03,380 --> 00:00:06,240
Nobody had mapped this part of the ocean floor before.
Three details cause most upload errors:
- The time uses a comma before the milliseconds, not a period.
- The numbering must run in order, starting at 1.
- YouTube accepts SRT but only the basic version: no styling markup is recognized, and the file must be plain UTF-8. If your text contains accented characters or curly quotes and the editor saved the file in another encoding, they can arrive scrambled.
A timing line is the only part that is hard to produce by hand. A 90-minute video needs well over a thousand of them, which is why nobody writes these manually.
Three Ways to Turn a Voiceover Into an SRT File
There are three realistic routes, and which one fits depends on what you have.
| You have | Method | What you do afterwards |
|---|---|---|
| Audio only | Speech recognition (Whisper-based tools and similar) | Proofread every line for names, numbers and terms |
| Audio and script | Forced alignment with an open-source aligner | Install and run it from the command line, then convert its output into SRT |
| Audio and script | Timing the script against the pauses in the recording | Spot-check a few sentences |
A note on cost, because it comes up constantly: Whisper itself is open source, so the model is not what costs money. The cost is in running it. Either you install it and have a computer that can handle a long file, or you pay a hosted service by the minute or by subscription. For a creator publishing several long videos a week, that is a recurring bill for a job where the text was already written.
Forced aligners are free but built for researchers. They tend to want a specific audio format, a cleaned-up text file and a terminal, and they rarely write SRT directly.
The third route is the practical one for scripted channels. A narrator, human or AI, leaves audible gaps between sentences and between thoughts. Those gaps are visible in the audio's loudness without anyone needing to understand a single word. Find the gaps, lay your known words across the stretches of speech between them, and you have timing that is anchored where the voice really paused.
One honest limit: the timing is tightest at the pauses. Inside a long unbroken stretch of speech, words are spread according to how long each is to say, so a caption can drift by a fraction of a second mid-sentence. For standard subtitles that is invisible. For word-by-word karaoke highlighting, you would want true word-level recognition.
What Breaks Sync, and How to Fix It
Script-based timing is dependable, but it fails in predictable ways. These are the ones worth checking before you blame the tool.
The voice does not say what the script says. The words are laid over the speech in the order the script gives them, so a skipped paragraph or an ad-libbed sentence pushes everything after it out of place, sometimes for the rest of the file. A text-to-speech voiceover reads word for word, which is why this method suits AI narration. If you recorded yourself and went off script, edit the script to match what you said before timing it.
The script contains words nobody speaks. Headings, section labels, [PAUSE] markers, and notes to yourself will be timed as though they were read aloud. Paste only the narration.
Music or sound effects sit under the voice. Pause detection reads loudness. If a music bed never drops below the speech level, there are no quiet gaps to find. Export the voice track on its own, before music is mixed in, and use that file for timing.
The file is several clips joined together. A very long video is easier to time in parts, one per chapter or section, with the matching slice of script each time. It also keeps any mismatch contained to one section instead of the whole file. If your chapters need timestamps too, how to add chapters to YouTube videos shows the format.
Whatever tool you use, spot-check three places before you upload: the first caption, one in the middle, and the last. If all three line up with the audio, the whole file almost always does. If the last one is early or late, find where the script and voice parted ways.
How to Upload an SRT to YouTube
Once you have the file, adding it takes a minute:
- Open the video in YouTube Studio and go to Subtitles.
- Choose your language and click Add.
- Select Upload file, then pick With timing, since an SRT carries its own timestamps.
- Choose your
.srtfile and click Save.
Video editors take the same file. Premiere, DaVinci Resolve and CapCut can all import an SRT as a caption track, which is useful if you want to style the captions yourself and burn them in.
A Free Tool That Does the Script-Plus-Voiceover Version
If you have the script and the voiceover, AutoTube.pro has a free script to SRT tool that does exactly the third route above, with no account.
The process has three steps:
- Paste the script. Only the words read aloud, in order.
- Choose the voiceover. An MP3, WAV or M4A that reads the script word for word.
- Download the SRT. Each caption is timed to your recording, pauses included.
Your audio is not uploaded. Your browser measures how loud each moment is, and only that loudness trace and the script are sent for timing, which keeps a large voiceover file on your own machine. The result shows the captions on screen before you download them, so you can read the first few lines against the audio right away. Very long recordings need a computer with plenty of memory, and the tool page lists the length limit.
It is a tool for one situation: you already have the words. If all you have is a recording with no script, you do need transcription, and the tool is not a substitute for it.
If your voiceover comes from a text-to-speech voice, that is the case where the match is guaranteed, because the voice reads exactly what you gave it. AutoTube.pro's AI voiceover produces that narration inside the same workflow as the script. The free SRT tool does not require it, though: it takes any audio file, so narration from anywhere works.
Get Your Captions Without Paying Per Minute
Captions are one of the last jobs before upload and one of the easiest to over-engineer. If the script exists, paste it into the free tools on AutoTube.pro, add the voiceover, and download an SRT you did not have to proofread for names. The same page lists the other free tools, including a thumbnail maker and a video length calculator.
FAQ
Can I convert a voiceover to an SRT file for free?
Yes. If you have the script, you can time it against the recording and skip transcription entirely, using a free script-to-SRT tool or an open-source forced aligner. If you only have audio, free speech-recognition software can transcribe it, but you will need to proofread the result.
Do I need Whisper if I already have the script?
No. Whisper is a speech-recognition model, built to work out what was said. When you already know what was said, you only need the timing, and that can be worked out from the pauses in the recording without recognizing any words.
Does YouTube accept SRT files?
Yes. YouTube accepts SubRip (.srt) among its basic caption formats, but only the basic version: styling markup is ignored and the file must be plain UTF-8. Upload it in YouTube Studio under Subtitles using the With timing option.
Why are my captions out of sync partway through?
Almost always because the voiceover and the script no longer match from some point onward, such as a skipped or ad-libbed sentence. Fix the script to match the audio, or time the video in sections so a mismatch stays local. Music left in the audio can cause the same symptom, because it hides the pauses used for timing.