Reading Graphics Text Base64 String From Premiere XML File

I'm building an app that reads the XML output of a Premiere project. For text graphics, Premiere appears to encode all the properties of the text (the string itself, font, size, etc.) as a JSON string stored in a base64-encoded string. Like this:

<parameter authoringApp="PremierePro">
										<parameterid>1</parameterid>
										<name>Source Text</name>
										<hash>4e081bff-5365-dc02-2a20-2ed4000001a4</hash>
										<value>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</value>
									</parameter>

The trouble is that this base64 string does not decode to a valid JSON string in ANY known string encoding format (I've tried 28 of them). The XML file itself is UTF-8 encoded. I found old documentation that said Premiere used UTF-16LE for these base64 strings, but that appears to be outdated:

https://community.adobe.com/t5/premiere-pro-discussions/encoding-charset-of-source-text-in-exported-graphicandtype-effect/td-p/9477477

(The example <value> string from this old thread *does* properly decode into a JSON string when the string encoding is set to UTF-16 Little Endian.)

There is something weird going on with this base64 string and I just need to know how Adobe encoded the data so that I can decode and read it. Thanks!