In my case, loading the entire file is loading a tiny bit more data, so this fallback doesn't hurt, but it's still annoying, and broke any hope I had of doing something more clever with the dataset.
I suspect this is where Windows backwards compatibility bites them a bit. I've got a very old tool [1] that uses WriteProcessMemory and CreateRemoteThread to create a thread in the command process that launched it to remotely change the directory in that process.
It works to this day, despite looking exactly like what malware would do. My tool is nothing in the grand scheme, but I suspect I'm not the only one doing these sort of shenanigans, and no doubt some big important app is doing it and can't be bothered to fix itself, so MS is stuck supporting it.
I ended up using the same basic layout for the database behind a little IP lookup tool I wrote to make lookups somewhat responsive from JavaScript [1]. It ends up working out pretty well.
WhisperX along with whisper-diarization, runs at something around 20x of real time on audio with a modern GPU, so for that part, you're looking at around $1 per twenty hours of content to run it on a g5.xlarge, not counting time to build up a node (or around 1/2 that for Spot prices, assuming you're much luckier than I am at getting stable spot instances these days).
You can short circuit that time to build up a node a bit with a prebaked AMI on AWS, but there's still some amount of time before a new node can start running at speed, around 10 minutes in my experience.
I haven't looked at this particular solution yet, but I really find the LLMs to be hit or miss at summarizing transcripts. Sometimes it's impressive, sometimes it's literally "informal conversation between multiple people about various topics"
The current directory is managed with SetCurrentDirectory/GetCurrentDirectory, however the cmd.exe command-line shell also stores the current directory for each drive in an environment variable like "=C:", and the CRT and shell hides all environment variables that start with a "=".
It gets mightily confused if these two concepts of current directory ever diverge.
I wanted to see if I could do the lookup work client-side, and also include some more metadata about cloud provider's IPs (region, service, etc), not that it's really better, just a toy idea I had.
https://github.com/seligman/podcast_to_text/blob/master/sear...
In my case, loading the entire file is loading a tiny bit more data, so this fallback doesn't hurt, but it's still annoying, and broke any hope I had of doing something more clever with the dataset.