Now in Premiere Pro Beta: Import Your Own Transcript

[TeresaDemel]

We have heard your feedback: Those of you that have your own transcripts, already complete with proper spelling and punctuation, or in a language we don’t support, don’t want to re-transcribe using Premiere’s Speech to Text feature just so you can use captioning and Text-Based Editing.

Captioning and Text-Based Editing requires Premiere Pro to have timecodes for every word. We are publishing a JSON spec format that supports word-level timecode for captioning and Text-Based Editing. Developers can now build support for exporting to this spec, or write scripts to translate from other transcript formats to this JSON spec. (JSON spec is attached to this post.)

Guidance for non-developers

While developers are adding support for this new format to their transcription tools, we’ve found that modern LLMs can do an excellent job of converting existing transcripts into this JSON format. Every LLM is different and their capabilities are always changing, so this advice is generalized and not guaranteed to work for every situation.

Use an LLM to convert your existing transcript:

1. Attach the JSON spec and your transcript file to the prompt
2. Ask the LLM to reference JSON spec and convert your transcript file into that JSON format, making sure to follow the spec and preserve the specific timecodes for each word or sentence. Alternatively, you may have better luck asking the LLM to write and execute a python script that translates the format for you. See below for an example of that prompt.
3. In Premiere Pro, load your source clip into the Source Monitor and open the Transcript tab of the Text panel
4. Click on the Overflow (…) menu, and choose Import > Import Static Transcript
5. Choose the JSON file that you created.
6. Your clip now has a transcript attached that can be used for Text-Based Editing or accurate caption creation.

Example prompt:

Here is spec file for a transcript format and a transcript that has been written in a different format. Write a python converter that converts transcripts in the given format into transcripts from the spec'ed format. Make the conversion as lossless and precise as possible. Make sure that you follow the spec to the letter and don't deviate from it. Document the Python code and make it work on single transcripts and folders of transcripts. After the python code is written execute it on the given transcript and return the converted transcript.

[Stan Jones]

@TeresaDemel,

Fabulous! Congrats to the team! I look forward to the experts weighing in on how their experiments go.

I do not have expertise in AI and large language model options. To further my education, I'm looking into free, local options. I'll head down that rabbit hole shortly (at least a little ways).

Stan

[_nicdean]

Super cool! Is there an update to ExtendScript API for options "Export transcript (json)..." or "Import static transcript"? That would be required for us to best make use of this new feature.

[mattchristensen]

@_nicdean not yet, but, of course we recognize how valuable that would be. Stay tuned!

[Stan Jones]

@_nicdean,

You had a great response in the Adobe Creative Cloud Developer Forums regarding future API/UXP options for captions. Not my expertise, but Teresa's post is directed in part to developers. I'm guessing that @bbb_999's response will be - after the initial "what cep can do" release.

In any event, I think it's time for you to post over there again and ask!

Stan

[_nicdean]

thanks @Stan Jones! I'll ping that thread.

[yuig22591562]

Great addition esp. for multilingual edits that are not properly managed yet but it immediately feels like the need to properly format a json first will act against the flow of i.e. a journalist and will still require an external tool that could be internalized. Pasting a text from the Windows buffer would be welcome. Also a simple notepad tool to keep raw texts as is or using them as source for a retranscription by Premiere, keeping  the text unchanged, using a user mark to point to the voice over as a reference for the timing.

[nbechere]

Thanks for these suggestions. The problem with Raw TXT files is that they don't contain any timecodes. So there's no (direct) way to use them for (Text Based) Editing.
In cases where you have generated a transcript in Premiere but have a better (e.g. manual) transcription at hand in the form of a TXT file - there's the existing option to "Import corrected transcript...". Through that option Premiere will try to infer the timecodes automatically by mapping the imported txt file the the timecodes from the internal transcription. 

But starting from scratch with a TXT file is sadly not possible yet. 

[DJP2014]

I take it this doesn't help translate Premiere's  .prtranscript file? 

Right now I'm trying to figure out a way for ChatGPT to generate a Word doc that highlights all the content I've used in a 4-min cutdown of a one-hour transcript. Without per word time-stamps I'm finding this very difficult.

[nbechere]

If I understand you correctly you might be in luck. Together with the option to import JSON we already offer export to  JSON now which writes the word level timecodes in human readable format. Does this help?

[Stan Jones]

@nbechere, That is truly massive. I overlooked @_nicdean's having noted that export option above.

Stan

[DJP2014]

Great news!

Word level timestamps open up lots of interesting possibilities for LLM.