MP3 to SRT: Convert MP3 to Subtitles with AI
Quick Answer
MP3 to SRT takes about a minute if you let a speech model do the timing. Upload the MP3 and you get a transcript where every line has a start and end time. Fix what it misheard. Download the .srt.
You can still do it by hand. You type each line and set its timing against a waveform in a free editor, which works fine but takes five to ten times the length of the audio. That's worth it when the clip is short or the audio is so rough the model keeps guessing.
What You Actually Get
An SRT is plain text and nothing else. Each subtitle is a cue made of a number, a timing line, the text and a blank line.
The timing line is hh:mm:ss,mmm --> hh:mm:ss,mmm. Hours, minutes and seconds, then a comma and three digits of milliseconds. You can't skip the comma. WebVTT uses a dot in the same spot. Most players tolerate a dot in SRT but a few strict parsers don't, so keep the comma.
The file knows nothing about your MP3. So the same SRT works with the MP3 or a video that carries the same audio or a re-encode of the recording. The times line up as long as the audio starts at the same moment.
The full spec is in the SRT format technical guide. For timestamp edge cases there's a separate post on SRT timestamp format examples.
Ready to try it with your own file?
Upload audio, review the timing, and export the format you need.
Two Cues From a Real File
1
00:00:01,200 --> 00:00:03,800
Welcome back to the show.
2
00:00:04,100 --> 00:00:07,450
Today we are talking about room acoustics
and why your mix sounds different in the car.
Cue 1 is on screen for 2.6 seconds. Cue 2 starts 300 ms after cue 1 ends, and that gap is normal. You want it there. Overlapping cues render differently from player to player, so keep a short gap. Netflix asks for at least two frames.
MP3 to SRT Converter: The AI Route
Use this route for podcasts, interviews, lectures, voice memos and songs.
- Open EasyLRC at easylrc.com/upload and sign in with Google. No card.
- Drop the MP3. WAV, M4A, AAC, OGG and Opus work too. So do MP4 and MOV if you only have the video.
- Wait. A three minute file comes back in about 10 to 20 seconds. The model transcribes the speech and times each line against the audio.
- Read the result top to bottom. Fix names, numbers and anything it misheard. Timing is usually fine. The words are where errors live.
- Export SRT.
The same job also exports LRC, WebVTT, TTML and ASS. If a client comes back next week wanting VTT for the website you don't re-run anything, you just pick a different export.
The free plan covers 5 minutes of audio a month with no watermark. Paid plans start at $5 a month and add word-level Enhanced LRC. It handles 99+ languages.
What to Check Before You Export
What the model gives you is a draft. Skip these checks and they tend to bite later.
- Two people talking at once. The model picks one voice and drops the other. Find those spots and split the cue, then label the speaker or cut the crosstalk. Nobody can read two overlapping lines anyway.
- Music under speech. Sung words get transcribed like speech. Intros and outros with no vocals can produce an empty cue or a junk word, so delete them or replace with
[music]. - Line length. Netflix's style guide says 42 characters per line and two lines per cue. BBC broadcast guidance is 37. A long sentence that came back as one cue should be split at a pause.
- Cue duration. Under a second is hard to read. Over six or seven seconds means the cue is carrying too much text and you should split it.
- The first cue. Check it starts where the speech actually starts. If your MP3 has leading silence that the video doesn't, every cue is off by the same amount.
The Manual Route: Subtitle Edit or Aegisub
Both editors are free and open source. Both load an MP3 directly.
Subtitle Edit runs on Windows 10 22H2 or newer, macOS 12 or newer and Linux via Flatpak. MIT license. It has a waveform view where you drag to set in and out points, and it writes SRT natively. There's also a speech to text option built on Whisper and similar engines, so you can get a rough pass locally and clean it up after.
Aegisub is cross platform and built for timing subtitles to audio. It leans toward styled ASS output but handles plain subtitles fine.
The manual loop is the same in either:
- Open the MP3.
- Type a line.
- Set its start and end on the waveform.
- Repeat.
- Save as SRT.
Budget five to ten times the audio length, so a 20 minute interview is a couple of hours. Which is why most people now generate first and edit second.
When You Want VTT or LRC Instead
SRT is the safe default. But two neighbours cover cases SRT doesn't.
| Format | Use it for | Timestamp | Extras |
|---|---|---|---|
| SRT | Video players, YouTube uploads, editing software | 00:01:02,500 | None. Plain text only. |
| WebVTT | HTML5 <track> on web pages, HLS streams | 00:01:02.500 | Positioning, styling, chapters |
| LRC | Music players, karaoke, lyric apps | [01:02.50] | Start time only, no end time |
Building a web player? Use VTT, since browsers won't load SRT in a <track> tag. The main change is dot versus comma and the SRT to WebVTT converter handles it.
If the MP3 is a song and the target is a music player you want LRC. The SRT to LRC converter drops the end times and rewrites the timestamps. Or just export LRC straight from the same transcription job and skip the conversion.
Common MP3 Problems
Lots of wrong words. Low bitrate MP3s, phone recordings and heavy reverb all raise the error rate. There's nothing to fix in the file. Plan for more review time and listen while you read.
Wrong language. Pick the right language on the upload page. The tool doesn't guess it, you choose it before the job runs. Mixed language audio is harder and will need a manual pass.
Cues drift against the video. The MP3 wasn't the same audio as the video. Maybe someone trimmed the intro or the video has a lead-in. Re-export the audio from the video or shift every cue by the offset. Both editors above can shift all times at once.
Player rejects the file. Check the comma in the timestamps and the blank line between cues. Then sequential numbering and UTF-8 encoding. Those four cover nearly every rejection.
Frequently Asked Questions
Can I convert MP3 to SRT for free
Yes. EasyLRC has a free plan. It covers 5 minutes of audio a month with no card and no watermark. For longer files you can go manual in Subtitle Edit or Aegisub. Both are free and open source.
Does the SRT reference the MP3 file
No. An SRT is only numbered cues with timestamps and text. It doesn't store the file name or any audio data, so the same SRT works with the MP3 or with a video that shares its audio.
How accurate is an AI generated SRT from an MP3
Timing is usually close. Words are where errors show up, mostly names and numbers and spots with music or crosstalk. Clean speech at a decent bitrate needs light review. Phone recordings need more.
Why does my SRT use a comma instead of a dot
SRT uses a comma before milliseconds and WebVTT uses a dot. They're different formats and a few strict SRT parsers reject a dot. If you need both, use a converter rather than find and replace.
Can I get LRC and SRT from the same MP3
Yes. One transcription job exports SRT, LRC, WebVTT, TTML and ASS. You don't re-upload or re-run anything to switch formats.
What if the MP3 is a song and not speech
It still works. The model transcribes sung lines and times them, though you should check sections with heavy backing vocals or effects. And if the target is a music player rather than a video, export LRC instead of SRT.
Ready to Create Your Own Enhanced LRC Files?
Try EasyLRC free—5 minutes of AI-powered word-level synchronization included.