Very interesting idea Scruffy.
As you commented, the cost is the big issue - the others are secondary.
DC++ is a detail that I am sure can be solved. It is probably the primary way you want to distribute everything.
Let me suggest that you take each reference recording and rename it, with the permission of the original author when known, to a particular standard. For example, start all filenames with something like "ATPS-" or "ATP_Scruffy-" (Alizée Translation Project by Scruffy). This way, all files of this type will group together. Also, everyone will know exactly which file is the reference. In addition, if others get involve, then they could hold the first 3 letters the same and change the end to reflect the different author or source.
For DALS, I assume that you are thinking of just using a version that has Alizée highlights. The full show would be too long (and expensive) and would be more apt to raise legal issues. This shows are already on DC++ so I doubt there would be any objection to making them available them there.
Again, interesting idea Scruffy.
Edit:
Looking at the automated route - I have looked into dictation programs before, and been told that most of these programs are very voice sensitive. In other words, they have to be trained to understand each voice. I will admit that I wonder if this is true for all, as programs like the default Windows dictation program does not seem to have this requirement. So, maybe there is something out there that would work that I have yet to stumble across.
Bottom line is that if you could get the video's language in writing, then the translation portion becomes much more manageable.