Question about extraction of subtitles and cue points for "USM" containers.
 

UPDATE: It seems, at least for some "USM" containers, patching in a modified video file may be possible, however further experimentation and research into compression used (if any) by Criware in their "USM" containers (currently testing on the 2012 title "The Amazing Spider-man").

UPDATE 2: Patching in videos didn't work, however it seems it may be possible to decode the cue points from the .usm container as they are stored in plain text, however some values are difficult to translate, I'll update in a few days most likely.

Hi all,

Over the past few months I've been researching the "USM" file format, with the end-goal of successfully encoding the videos within such a file, as most from older titles ("Batman: Arkham City", "Batman: Arkham Knight" & "Mad Max") can be encoded with very little perceivable loss in quality to the naked eye, with the result of significantly reduced file size. However, "USM" containers with cue points will soft-lock the title upon completion of the video file, when attempting to execute an encoded "USM", requiring either the end-user restart the title and skip the video file, or, (and is my current solution), extend the video file to a given length, with a disclaimer informing the end-user they must skip the video.

This issue has been documented in passing [1][2], however a solution hasn't yet been found (to my knowledge). If anyone has any insight into this file format, and extraction of cue points & subtitles, or might be able to point me in the right direction to sourcing the CRIWARE SDK and/or the AutoDesk Media & Entertainment Collection, I'd greatly appreciate it.

https://help.autodesk.com/view/SCLFR...itle_info_html
look this.

Thank you for linking this, it slipped my mind that this may be a relevant resource. I've already been through that particular set of documentation, to better understand the function of cue points, however it wasn't able to help me identify where to look within the compiled ".usm" for the relevant cue point information.

UPDATE 1: First-off thanks to some testing from users I contacted this project seems to be a success! Secondly, regarding subtitle meta-data extraction, the time-code for subtitle termination doesn't seem to be stored cleanly, whist the subtitle "begin" time-code is stored in a universal manner. Because of this it would require a user to listen through each sound file to determine when the subtitle should terminate, and I personally can't be bothered to do so. I'll probably research this further in the future but for now I've stopped examining this archive/container type, and I'm content with being able to gather the cue-points. Some time in the future I'll make a guide for anyone else looking to do so, and I'll try to possibly automate the process, as it's a rather convoluted process, and is not friendly to user's unfamiliar with hex-code.

-----------------------------------------------------------------------------------------------------------------

After extensive testing and a few sleepless nights I believe I've managed to decode the cue points in their entirety. for those of you with the title "Batman: Arkham City" I have this proof of concept (I would have gotten this far earlier, but I didn't anticipate the formatting of a given cue/trigger being so different to the examples provided in the now deleted 2nd post). It's the first cut-scene of the formerly mentioned title. There are elements I'm still working on, with their being some difficult to find cues referenced at the start of the video but as a grounds to build off of it seems to be successful after testing on my machine.

If someone else could test this and report back whether or not the title freezes up or resumes to game-play after the conclusion of the cut-scene I'd appreciate it greatly.

Should my own testing be successful I'll move onto the 2nd part of my project, decoding the subtitles (stored as plain-text) and their associated timestamps. This theoretically should be easy to do, provided the subtitle meta-data is encoded in the same format as the cut-scene triggers/cues.

Yo man, you've been dealing with .usm files.

Can you explain a little about cue points in a simple manner?

Cue points are essentially a set of instructions executed at a given point (or duration) within the runtime of a given video file. These can, for example, include instructions when to end rendering and when to activate controller rumble features and at what intensity to do so.

Thankfully not all titles that utilises CRIWARE's container format include cue points, as some engines allow for such instructions to be executed in-engine. It can also be a developers choice to include/exclude the use of cue points, for reasons I'm not yet clear on. Deciphering these cue-points can be trivial (for titles such as "Batman: Arkham City"), but can also be trickier for other titles, as there is no universal set of instructions, and even formatting of the cue points can change from title to title. So far I've had success with each title I've attempted this with, however more testing is required to determine whether these methods can be applied universally (depending on how the ".usm" is loaded by a given title (not engine-specific) into memory, and a guide can be published, as some encoded videos will not terminate correctly, even if the methodology is identical to a working, encoded ".usm".

I'm also reluctant to release a guide just to spite some users over at the Xentax forums who were really unhelpful buggers, but were still expecting me to educate them should I be successful.

Keep digging man these .usm files are kinda pain xD.
Anyway do you have a fast contact?

Concentrate on having fun with your experiments. Then releasing your results (which you should do) or keeping them for yourself won't matter anymore. Just like it makes no sense to not release the source of a piece of software that is freely distributable anyway.

I'm not sure what you mean by this, was pulling an all-nighter packing a handful of emulated releases, so I'm a bit tired, brains fried rip :o

Honestly I'll probably release my findings in depth eventually, but just haven't the energy to write things up formally, it was a bit draining thinking I found a universal solution when I had actually just gotten lucky with an aspect of the process. On top of this I only really want to release a guide when I completely understand the container type, with subtitle meta-data (time-stamps), and attempting to identify whether or not a given track can have its track-number identified in a universal way (some tracks are labeled outside of the standard 0 - 11 numbering system). For now I've managed to create a proof of concept and apply my technique only to one title fully.