Create captions - 'minimum duration in seconds' being completely ignored

Report · Aug 04, 2023

Issue: When generating captions from a transcript, the 'Minimum durations in seconds' setting is ignored.~

Version: 23.5.0 (Build 56)
OS: Windows 10
System Info:
-i9 10290x
-RTX 3070, version 531.79
-64GB, 3200mhz RAM

-Software/footage is on a 1TB NVME

Video format: H264, 25Mbps, 1920x1080, exported from Premiere Pro.

Workflow details:

Just the exported video on its own sequence.

Steps to reproduce:
- Add footage to a sequence, and generate a transcript.

- Create captions from this transcript, set a minimum duration of 3 seconds, set double lines.

- Generate captions

- Review captions, notice that they do not adhere to the minimum duration, nor double line request.

Expected result:

Captions that remain on screen for 3 seconds, and double lined where necessary.

Actual result:

Several single line captions, some lasting less than one second.

Attached:
My caption settings used
My transcript that it's generating from
My captions that are being generated

Further note:
Not an isolated issue - several editors I know have this same result.
From what I can tell, Caption settings seem to be overridden by the punctuation in the transcript - because the transcript generated with erroneous full stops/periods - it generates a new caption. This is despite the fact they could all fit in a one or two line caption, within the length/duration set.

Further testing:

Even setting the max length to the highest setting, it will still use the period/full stop as a point to make a new caption (i.e. makes no difference to these initial few lines). Setting the duration appears to also have little to no effect. It still seemingly wants to use Periods as a point to generate a new caption, regardless of intention.

Report · Aug 05, 2023

@Hypohamish,

Thanks for raising this issue, and for including so much detail. Here's another report:

https://community.adobe.com/t5/premiere-pro-discussions/create-captions-from-transcript-ignores-mini...

I did a bit of testing (using Release 23.5.0), and I cannot make sense of some of the logic here. I am most interested in the interaction between the various options. I tested only the Subtitle format. There may be substantial differences in how closed caption formats operate. And I ignored the Gap option. First, I have a few general comments.

The basic “Create Captions” Dialogue is unchanged from the early 15.x release, except Max and Min character length.

The options in this dialogue include preset options, Style, Maximum length in characters (default 42, max 72, min 7), Minimum duration in seconds (default 3, min 1.2, max 6), Gap between captions (in frames: default 0, max 10), and Single/Double Lines.

Information on these options is in this help file, but the information is limited.
https://helpx.adobe.com/premiere-pro/using/speech-to-text.html

It is important to keep in mind that each option cannot be primary. For example, if we specify a line length of 42 characters and single line, many caption segments with more characters must move to a second line, or have short display times.

But, as you point out, sometimes the process produces very short captions for less than the selected minimum time.

Maximum Characters per line appears to be primary, meaning that PR will respect this over any other setting. And single vs double may be second in line. But captions are in a word wrap format, and even though PR may only add an actual line break at a particular location, larger font size or narrower bounding box limits may result in the display showing more lines.

But no matter the minimum duration setting, caption segments were often broken as short as less than one second even though they did not exceed the character max, with and without considering single vs double line.

Yes, often periods can result in breaking caption segments, but I also saw instances where such punctuation did not break a segment.

Stan

Report · Aug 06, 2023

Thanks Stan for your exhaustive findings and research there.

As you say, the further I look into this rabbit hole, the worse it seemingly gets. I've had further erreoneously generated captions which, as mentioned, don't even have the justification of punctuation as to why they're generating so stupidly.

Should someone from the higher Adobe help make it into here, I've attached further examples below.

Despite captions being on default settings (length 42 character, duration 3 seconds), it has broken one line into two individual captions. Note there isn't even a pause in the audio, the speaker literally spoke a fluid sentence here without interruption.

This is the transcript the above caption was generated from.

This seriously needs to be better. I can only assume people haven't posted about this en-masse in the past is either because people are editing the transcript and then auto-generating captions and calling it a day, or people are creating more bespoke content with captions and therefore don't mind playing around with spacing etc.

Report · Sep 13, 2023

My coworker and I have been experiencing the exact same issue. It is very frustrating as we sometimes need to make significant edits to the captions even after having cleaned up the transcript and carefully selected our settings. Adobe, please address this.