How to Create an SRT File from Audio and Check the Timing

Published on October 15, 2026 by Steven Jones

To create an SRT file from audio, transcribe the recording with timestamps, export the subtitles as .srt, then play the file alongside the original recording to check when each caption appears and disappears. A correct transcript is only the starting point: badly placed subtitle boundaries can make accurate words difficult to follow.

You can produce the first timed draft with the DIY AI Studio speech-to-text tool, or write a short SRT manually. This walkthrough covers both routes, with a four-cue example built from original demonstration audio, a timing checklist, and fixes for subtitles that start late or drift out of sync.

Create and check an SRT file in five steps

1. Use the final audio. Remove unwanted intros, silences or edits before generating timestamps. A subtitle track made for an earlier cut can be perfectly formatted and still be wrong.
2. Transcribe with timestamps. Upload the audio to a tool that produces timed segments, not just a plain TXT transcript.
3. Export SRT. Save the timed captions as a .srt file, preserving the numbering, timestamp arrows and blank lines.
4. Edit the subtitle cues. Correct names, punctuation, breaks and each start/end time against the audio.
5. Test playback. Load the SRT with the finished media, inspect the first and last cues, and watch at least one middle section.

---

---

A checked SRT example you can copy

We created a 14.834-second test recording from an original four-sentence script, spoken by a locally generated eSpeak British English voice. We entered the words manually from that script, not from an AI transcription service. We measured audible speech boundaries against the waveform, and checked the resulting SRT in FFplay against the matching media. This demonstrates subtitle construction and timing, not the transcription accuracy of DIY AI Studio.

Copy the following into a plain-text editor to inspect or adapt it:

1
00:00:00,313 --> 00:00:03,093
This is a short subtitle
timing test.

2
00:00:03,860 --> 00:00:06,717
Start the caption when
the voice begins.

3
00:00:07,498 --> 00:00:10,418
Leave a short gap
when the speaker pauses.

4
00:00:11,216 --> 00:00:13,883
Check the final cue
before you publish.

For this file, we verified UTF-8 encoding, consecutive cue numbers from 1 to 4, increasing timestamps, no unintended overlaps and an end time inside the audio’s duration. At playback, the captions followed the spoken phrases, with empty intervals left for the pauses. The gaps are intentional: extending a caption across silence would make it appear that the speaker was still talking.

How to transcribe audio and export SRT in DIY AI Studio

1. Open Speech to Text in DIY AI Studio and sign in.
2. Choose an MP3 or PCM WAV recording that you have permission to process. Set the recording language to match the speech.
3. Confirm the permission checkbox, review the credit price shown for the recording, then select Transcribe audio.
4. When processing finishes, review the transcript against the recording, paying particular attention to names, numbers and words near pauses.
5. Use the result’s export options to download the SRT subtitle file. Open that downloaded file in a subtitle or text editor for the final timing and wording corrections.

Important Studio export limitation: the TXT download uses the text edited in the browser, but the SRT and VTT downloads use the original timed segments. Do not assume a correction made in the editable transcript has also changed the exported SRT. Check and correct the SRT itself before publishing.

If your goal is simply a readable document with no subtitle timing, follow the separate MP3-to-text guide. You do not need an SRT file for an ordinary transcript.

How to make an SRT file manually

For a very short clip, manual creation can be quicker than setting up a transcription job. In Windows Notepad, paste the sample above, replace the words and timestamps, select File > Save as, set Save as type to All files, choose UTF-8, and name the file my-video.srt. On a Mac, switch TextEdit to plain-text mode before saving. Confirm the file did not become my-video.srt.txt.

Each cue has a number, a start and end timestamp, one or two lines of text and a blank line before the next cue. Time uses HH:MM:SS,mmm: 00:01:02,500 means one minute, two seconds and 500 milliseconds. Use a comma for milliseconds and the exact --> separator between times, not a full stop or a typographic arrow.

The Library of Congress description of the SubRip format documents this numbered, timestamped structure. SRT has no single mandatory character encoding, but saving in UTF-8 is a sensible compatibility choice, especially where names or other languages include accented characters.

Set cue boundaries and line breaks around the speech

Start a caption close to the first audible word and end it shortly after the last. Do not move every boundary to the nearest whole second. A 300-millisecond error is especially noticeable at a quick speaker change or after a sharp pause.

Break lines at natural phrases rather than splitting a name, number or verb from its subject. Aim for no more than two lines and roughly 32 to 42 characters per line as a starting check, then review what the destination player actually displays. These are editing guidelines, not a universal technical limit. A line that looks neat in a text editor can still run off the screen in a mobile video layout.

Also consider how long the text stays visible. If a cue contains 70 characters but lasts only two seconds, a viewer would need to process 35 characters per second. The correct fix may be to shorten the wording, split the cue at a genuine pause or alter the edit, rather than leave a technically valid but unreadable subtitle.

Fix subtitles that have a constant offset or gradually drift

What you see	Likely issue	Best correction
Every cue appears about half a second late	Constant offset, perhaps from an added or removed intro	Shift all start and end times earlier by 500 milliseconds, without creating negative values.
The first cue matches but later cues become increasingly late	Drift caused by differing source duration, playback speed or a changed edit	Compare an early and late speech marker, check that the media versions match, then realign or regenerate the timed track.
Only one line is out of sync	Bad segmentation around a pause or speaker change	Adjust that cue and neighbouring boundaries rather than moving the whole file.
The captions are aligned to the audio, but the video lip-sync is wrong	Audio and picture disagree	Fix the media sync first. Rewriting the SRT will not repair the underlying video.

Use two checkpoints before applying a global fix. If a recognisable word is 400 milliseconds late at 30 seconds and still 400 milliseconds late near the end, a simple offset is appropriate. If it is correct near the start but three seconds late near the end, shifting everything will only relocate the error. Re-sync against the exact final media file instead. This is especially relevant after trimming podcasts, speeding up recordings or replacing an exported video with another version.

If poor sound is causing missing or invented words, address the recording before spending time on subtitle boundaries. Our guide to transcribing noisy audio covers that separate problem.

Your search, your sources

Make DIY AI a preferred source

See more of our reporting in Google Top Stories, AI Overviews and AI Mode.

Add DIY AI on Google

Import the SRT and verify playback before publishing

For a video, place my-video.srt beside my-video.mp4 so a compatible player can find it, or manually load the subtitle file. In VLC, use Subtitle > Add Subtitle File if the subtitles do not appear automatically. Check that captions are enabled. Not every audio-only player displays an SRT track, so use a video preview or a subtitle editor with an audio waveform when working from a standalone MP3.

Scrub to the beginning, one pause in the middle and the final sentence. Check that no caption arrives before its words, persists over the next speaker or disappears halfway through a phrase. Then watch the finished sequence at normal playback speed. A file can pass a format check and still be uncomfortable to read.

Finally, make sure the subtitles are attached to the exact media version you intend to upload. If you change the audio or video edit afterwards, repeat the timing check. Keep the original recording, corrected SRT and published media together so you can identify which version the subtitles belong to.

Frequently asked questions

Can I make an SRT file from an MP3 without video?

Yes. Speech recognition can generate timed segments directly from the audio. You only need a compatible video preview or subtitle editor when you want to see the text appear in sync.

Can I convert a plain transcript into SRT by renaming it?

No. A plain transcript has no cue numbers or timed boundaries. You must divide it into captions and assign start and end times, either manually or with a timing tool.

Does SRT contain the audio?

No. It is a separate plain-text subtitle file. Keep the corresponding audio or video alongside it for testing, and use the caption import option required by your publishing platform.