Audio to subtitles

Audio to SRT: subtitles from a recording.

An SRT is not something audio converts into — it is a transcript with a start and end time on every line, so making one means transcribing the recording. Open the audio file (MP3, WAV, M4A, FLAC and the rest) in a speech-to-text subtitle tool, let it write timed lines, correct the words it got wrong, and export SubRip (.srt). In suubbs that is drop the file in, press Generate, export — on your own Mac, with nothing uploaded. The free browser version does the first 60 seconds; the Mac app does the whole recording.

Get the Mac app — $9.99 → Try it free in your browser
Step by step

From an audio file to an .srt in five steps.

The same route as a video, minus the picture.

  1. Open the audio file. In the Mac app, add it to the library or drop it on the window — the file picker lists MP3, M4A, WAV, AAC, FLAC, Ogg, Opus and AIFF alongside video. In the browser studio, drop it on the player; the browser decodes it, so any format your browser can play will work.
  2. Set the spoken language. Open ⋯ → Transcription settings. The Mac app can detect the language; the browser version asks you to pick it. Telling Whisper the wrong language tends to produce a translation rather than an error, so it is worth the two seconds.
  3. Give it the names (Mac app). Type guests' names, product names and jargon into "Names and jargon". It biases the transcript toward your spellings, which is where most corrections in a podcast would otherwise go.
  4. Generate, then read it through. Click a line to correct it. The timeline draws the recording's waveform, so you can see where the pauses are and drag a cue edge that is wrong.
  5. Export SubRip (.srt). Or WebVTT for a web player, or word-level JSON if you want every word's timing for something else.

With no picture to measure, the studio lays the subtitles out for a widescreen frame — up to 42 characters a line, two lines a cue. If the subtitles are going over vertical video, switch Timing & layout → Line layout to "Vertical / social", which allows 22 characters and three lines, before you export.

What the file contains

Timed text, and nothing tied to a picture.

Which is what makes an audio SRT useful in more than one place.

An SRT made from audio is identical in shape to one made from video: a numbered list of cues, each with a start and end time down to the millisecond and one or two lines of text.

1
00:00:00,480 --> 00:00:03,120
Welcome back to the show.

2
00:00:03,400 --> 00:00:06,950
Today we're talking about
how to record in a small room.

Times are counted from the start of the audio file. That one fact decides whether the file lands correctly wherever you use it — see the section on sync below. The SRT generator guide covers the format line by line.

What people make audio SRTs for

Drop in the recording. Export the SRT.

[nothing uploaded]

The trap

When the audio and the video do not start together.

The SRT is timed to the file you transcribed, not to the edit you will put it on.

If you transcribe the raw recording and the finished video starts with a five-second intro, every line will appear five seconds early. If the edit cut a section out of the middle, everything after the cut drifts. The subtitles are not wrong; they are timed to a different file.

Three ways to avoid it, best first:

  1. Transcribe the final mix. Export the audio from your finished edit — or open the finished video itself — and generate from that. The timings then match the thing people watch.
  2. Shift everything at once. If the only difference is a fixed offset, open Timing & layout and use "Sync — every cue at once", which moves every cue up to three seconds earlier or later and changes the exported timings themselves.
  3. Fix it in the editor. Import the SRT onto the timeline and slide the track to the audio's position there.

If the subtitles start right and get steadily worse, the problem is not the start point; the subtitle sync guide walks through telling the kinds of drift apart.

Getting a clean transcript

What the recording does to accuracy.

Speech models are good. Bad audio is still bad audio.

Be clear what you need

Subtitles, or just a transcript?

They are different deliverables, and suubbs makes one of them.

suubbs is a subtitle studio. It exports SRT, WebVTT, styled ASS and word-level JSON — all of them timed — and it does not export a plain-text transcript. If what you want is a document to read, edit or quote from, such as meeting notes or show notes, an SRT is the wrong shape: it is broken into short timed lines with numbers between them. A transcription app built for documents will serve that better; the suubbs vs MacWhisper page is honest about where that line falls.

If you want captions — text that appears in time with the speech, on a video, a player or a social clip — the timed file is the point, and that is what this is for. Burning the captions into the picture instead is covered in burn subtitles into a video.

The details

What people ask next.

Including the limits.

Can I convert an MP3 to SRT?

Not by conversion — an MP3 contains sound and an SRT contains timed text, so the words have to be transcribed. Open the MP3 in a speech-to-text subtitle tool such as suubbs, generate the subtitles, correct them, and export SubRip (.srt).

Which audio formats work?

The Mac app's picker accepts MP3, M4A, WAV, AAC, FLAC, Ogg, Opus and AIFF. It decodes MP3, WAV, FLAC, AIFF, Ogg Vorbis and AAC or ALAC itself; Opus and rarer formats need an ffmpeg already installed on the Mac, which the app falls back to. The browser version decodes whatever your browser can play.

Is there a length limit?

In the free browser version, the first 60 seconds are transcribed. The Mac app transcribes the whole recording, however long, at no per-minute cost.

Is my audio uploaded anywhere?

No. In both versions the speech model runs on your own computer, and the recording is read from your disk and never sent anywhere. The offline subtitle generator page lists everything the Mac app does download.

Why are my subtitles a few seconds early or late on the video?

The SRT is timed from the start of the audio file you transcribed. If the video has an intro or was cut differently, shift every cue at once in Timing & layout, or better, transcribe the audio of the finished edit.

Can it handle more than one speaker?

It transcribes everyone, but it does not label who is speaking. Overlapping speech is where words get dropped, so check those passages; a larger model helps.

Can I get a plain text transcript instead?

Not directly — suubbs exports timed subtitle formats (SRT, VTT, ASS and word-level JSON), not a .txt transcript. If you want a document rather than captions, a transcription app is the better tool.

Can it translate the audio?

Whisper can write English subtitles straight from speech in another language: turn on Translate to English in ⋯ → Transcription settings.

Keep reading

Related

Any length of recording. On your own Mac.

[applause]