What can be transcribed
Generate Captions can only read clips that play recorded sound. Before a run starts, Frame Job works out which clips those are. Use this page when a clip gets left out, or when a long project stops a run.
- Menu
- File ▸ Generate Captions…
- Agent tools
transcribe
Which clips can be transcribed
Section titled “Which clips can be transcribed”A clip can be transcribed when all of these are true:
- It plays a movie or sound file that has an audio track.
- It plays forwards.
- Its audio is on.
- Its file is present.
File ▸ Generate Captions… is greyed out when no clip on the open timeline plays sound from a media file, forwards, with its audio on.
Whole timeline and In/out range read only clips you can hear. A clip that’s muted, turned down to silence, or on a muted or soloed-out track is left out, so a muted camera doesn’t repeat the words of the microphone beside it. A clip you can hear anywhere is read whole.
To caption a muted clip, select it and choose Selected clips, which reads every selected clip, heard or not.
Why a clip is skipped
Section titled “Why a clip is skipped”A clip that can’t be read isn’t part of the run. In the Generate Captions sheet it doesn’t count towards the summary line, so if the sheet says 2 clips and you expected 3, the third is here. The sheet doesn’t list these clips one by one. If it leaves nothing at all, it reads “Nothing to transcribe”. If there are clips to read but none you can hear, it reads “Nothing heard can be transcribed. Select a clip.”
An agent that names the clips it wants gets the reason back for each one, in these words:
| Reason | What it means | What to do |
|---|---|---|
| its source is not a media file, so it has no speech | It’s a text, shape, still, image, graphic, adjustment or nest clip. | For a nest, open it and caption inside it. |
| it plays in reverse, so its speech cannot be transcribed | The words would come out backwards. | Only a clip playing forwards can be read. See Reverse a clip. |
| its media has no audio track | The file has no sound. | Nothing to transcribe here. |
| its audio is switched off | For example, the picture half of a clip after Detach Audio. | After Detach Audio, the audio-only clip below it holds the sound and is read instead. Clip ▸ Attach Audio joins the two again. |
| its media file is missing | Frame Job can’t find the file. | Relink it. |
| its speed ramps, and no speed map was read for it | Frame Job couldn’t work out where the ramp puts the words, so it won’t guess. | Remove the ramp if you need this clip captioned. |
| its speed or length is degenerate | The clip has no usable speed or no length. | Add the clip to the timeline again. |
A clip that’s read but holds no words shows up afterwards, in the run report, as “no speech was recognised in it”.
Retimed clips land correctly
Section titled “Retimed clips land correctly”On a sped-up, slowed or speed-ramped clip, each word lands where it’s heard on the timeline, not where it sits in the file. Transcribing a ramped clip places each word where the ramp plays it.
Captions keep following afterwards. Change the speed later and they move with the speech. See Captions follow a speed change.
The 10-hour and 4,096-clip limits
Section titled “The 10-hour and 4,096-clip limits”AssemblyAI takes at most 10 hours per file. What counts is the length of the whole file, not the part your clip uses, because the whole file is what’s sent. Frame Job won’t extract or send a longer file, and its clips are listed in the run report with ““name” is longer than AssemblyAI’s 10-hour limit for one file.” The other files in the run carry on.
On this Mac there’s no limit on length. There is a limit on time: each clip gets four times the length of its audio, and at least two minutes. A clip still going after that is stopped and reported with “Transcribing “name” took too long and was stopped.”
A project holds at most 4,096 clips. When a run’s captions wouldn’t fit, none of them are placed and the project doesn’t change. The card says “5000 captions is more than this project has room for: 900 more clips would fit, out of 4096.” Frame Job can only count captions once the words are in, so this comes at the end of the run. With Replace ticked, the clips on the tracks it would remove don’t count against you.
Every caption or page is one clip, so word styles usually use more clips than Subtitles, and One Word uses one clip for every word. Limits lists the rest of the project’s limits.