Nice, been working on this post and project for a bit over a month and a half, with the big move in between of course. Glad it finally went live.
I built an open source image description service which will eventually get its own NVDA addon, and is of course hosted on fly.io. fly.io/blog/llm-image-descript…
Wanted to gut-check a project idea to see if anyone other than me would find value in it.
I have NVDA running across a handful of machines. I also, over the years, have done lots of subtle config changes, installed a handful of addons I prefer, etc.
Was thinking it'd be nice to have an NVDA sync service. Create an account, link your NVDA installations to it. Optionally mark specific config keys/addons to not be synced from specific installs, but otherwise keep everything synced everywhere else.
Add an ollama image recognition endpoint as an optional, or possibly limited-use, service addon for anyone wanting that capability but unable to run/host it locally. That'd probably be a paid feature, but config/addon sync is probably easy enough to do for free. Hell, maybe even train a chatbot with NVDA docs, GitHub issues, etc. and see if it can handle support requests.
Creating/hosting that all should be easy enough, sans the AI bits which would eventually cost but I could probably prototype something for free, though I'd eventually need to monetize it even if only to pay its own hosting costs if I'm ever unable to cover those myself. This feels like something I'd use, but I'd rather not spend the time if I'm the only one.
Does anything like this exist? I regularly find myself typing the same Slack bot commands, same shell commands, etc. Looking for a Windows app with a global hotkey which, when pressed, pops up an autocomplete box with titles like "Lock gateway," "Lock dev host", "Strip ANSI codes", etc. Each title is associated with text which, when the title is selected, gets pasted in. The autocomplete box also has an item for adding new snippets.
It'd probably be easy enough to write but it seems like it ought to exist already.
Windows Terminal has a feature where ctrl-clicking a link opens it in your browser. Anyone know if there is a screen-reader-accessible or keyboard-driven way to do this?
We use Teleport at work, which regularly prompts me to visit a long-ass URL to log in. The only way to do that is to use NVDA's review cursor to find the "http", set a start marker, arrow by character past the URL (which ends in a UUID so that's always fun since I have to review past 36 characters for that), set an end marker and copy the link. Rinse/repeat for any link dumped to the terminal. Takes me 20 or so seconds when all is said and done and I wish there was a faster way.
Wow, the #HomeAssistant folks weren't kidding at all when they said this would be their year of the voice.
Each new release seems to bring exciting new voice features. The last few have made it possible to run everything from speech-to-text and better text-to-speech locally, or to fall back to the cloud for TTS if you'd prefer.
Digging in a bit more, it looks like they're building out a standard protocol called Wyoming to make everything, including intent handlers/skills, separate services that you can run locally and use across assistants.
They're selling a few $13 USB C-powered mics. Once this gets wake word detection, apparently planned for later this year, I'm all in.
So I pulled the trigger and bought myself a Komplete A61, stand, and sustain pedal. Also picked up the piano by ear audio courses. I played piano for years as a kid and enjoyed messing around a bit as an adult, so hopefully these will be a good refresher.
Maybe I'm getting a bit ahead of myself, but are there any accessible live performance setups for the Komplete ecosystem? Or do folks just run Reaper even when playing live? I did a few jams in college and really enjoyed them, but that was 2002, and hauling my multiple-10-pounds of gear back and forth on the bus was painful enough that I only did it a handful of times. Might be more interesting with more portable gear. I guess remote jams are a thing now too
Haven't seen anything about this on the fediverse yet, but the PG13 folks will be streaming games from the recent Games for Blind Gamers jam in a little over an hour. Check it out--their streams are always a fun time. https://www.youtube.com/watch?v=tDBjPfOhFUo Enter a title
Twitter fired its accessibility team. Reddit apparently never cared about accessibility to begin with, and is torpedoing the lifeline lots of folks depended on to use it. And now Discord's username changes are apparently forcing folks through inaccessible hcaptchas. I hate to suggest that we need to collectively migrate back to hosting our own shit in order to have semi-stable communities where accessibility is a core tenant of those communities, but the alternative is looking pretty bleak and the tech bros don't care.
Excuse me while I alt-tab into Discord to figure out if I'm finally on the chopping block^H^H^H^Hup for username reassignment.
It finally happened. Got my Discord username modal today. And there's no audio CAPTCHA option, only the accessibility cookie that I guess requires me to compromise browser security. Fuck that shit with a hot rusty spork, is what I say.
I clicked the "I'll do this later" button, but if they haven't fixed this issue by the point where they force it, I'll be leaving Discord on principle until they do.
I'm a lot less butthurt about the fact that Discord's response to me submitting an amazing resume and cover letter for a senior accessibility developer position several years back, in the midst of a huge Texas snow storm, was a canned automated rejection a month later with no interview. I pulled out all the stops--mentioned that I'd written an Android screen reader along with mini screen readers for several UI toolkits, was familiar with accessibility in multiple web frameworks, etc. I should have at least gotten an interview. But the message is as clear today as it was back then: we don't really want disabled users on our platform.
So if it comes down to it, I'll be figuring out how to make Matrix work better for me and mine. The writing's pretty thoroughly on the wall for Twitter and Reddit. Now it looks to be there for Discord as well.
Holy fuck do I hate GitHub CI so much. Just tried updating godot-tts and it vomited 4 errors. Basically what that means for me as a blind screen reader user is digging into the raw logs for each failed job, hitting pageup half a dozen times, and trying to find whatever error caused the job to fail amidst thousands and thousands of lines of inconsistently-formatted spew. If I'm lucky I can search for "error" but it's usually far from that simple.
I'm out. Today I'm resentful that this has been a thing multiple screen reader users have struggled with for years, and a multi-billion-dollar company can't be bothered to fix it. Just because someone tells me I have to try harder to do what able-bodieds take for granted doesn't mean I actually do.
Hopefully one of the other folks wanting this project steps up and fixes the errors. To be clear, I don't resent my users. I resent how this is a tool that piles of open source projects use, the UX for getting at errors probably takes me an order of magnitude longer to use, and I'm supposed to just take that in stride and try harder. Naw, not today.
I've never messed with GPUs under Linux. Anyone have tips on how to stress-test one, or otherwise put it through its paces, that aren't gaming or mining crypto?
I realize the latter might be OK if I did it only in the context of a stress test and for a limited time, but aside from wanting to stay away from crypto on principle, this is for work and we have systems actively scanning for and flagging it.
Context is that I'm stress-testing our new GPU offering at work to find bugs. It doesn't necessarily have to be a stress-test, but I'm trying to actively break things in a way that won't trigger our scans but I don't know the first thing about this stuff.
And FWIW we've already generally run LLMs and such, so the lowest hanging fruit is already picked.
Well that was annoying. Last weekend Friendica started doing this thing where every post, including threaded replies, had an h5. Literally started happening within an hour or so of me posting about how accessibility had been improved. This made it incredibly difficult to keep up since I essentially had to read my entire feed, comments and all. And given Friendica's default mode of pulling in thread-starters that folks reply to and grabbing whatever thread context it can, I was about done.
Turns out Firefox nightly is doing this thing where it opens my history sidebar. Or maybe I am somehow and just don't know it. If as a screen reader user you've ever had to maximize a page to get all your controls back, it's like that, except maximizing made no difference.
Anyhow, it's fixed now. I wish I knew why this sidebar kept popping open, even though I have it set to not open unless I do it myself.
Set up an ollama server on my home network a while back and it's kind of magical to have all this capability just available to me anywhere (thanks to my Wireguard VPN, at least.) It's CPU-bound so a bit slow, but I've used it for everything from cleaning up writing to generating fanfic for my cat.
Since I set it up a couple months back, they've added support for image classifiers and OpenAI compatibility. Sorely tempted to buy an external GPU to speed things up. I do have access to cloud-hosted GPU hardware through work, but there's kind of a cool factor to running the AI on my assorted pile of hardware in my home. Wonder if anyone has integrated it with Home Assistant's voice assistants yet.
Never thought I'd be that guy raving about how great his company is, but...
Just got out of a nearly hour-long work meeting with the folks we're contracting with about audio describing the visual aspects of our town halls. Demos, presentations, I'll have my own personal audio describer in a shared Slack channel and huddle describing things out loud and answering my questions. We're also going to hash out presentation guidelines so folks doing demos do a few things to make everyone's lives easier, including watchers on phones where they can't see commands or screenshares.
Mind absolutely blown. Do FAAMG-scale companies do this for their employees, or just mine? This was a unique situation for everyone in the meeting, including the multiple companies contracting and subcontracting on this, so I suspect I just hit the jackpot and no one else does this.
I'm truly spoiled. Fly.io set the "cares about employees" bar pretty high and if I ever lose this job I'm going to be hard-pressed working for anyone else.
Never thought I'd be one of those folks going on about how great their company is on social media but here I am.
We have a web-based terminal that pre-installs/auths our CLI so you can upload code and launch apps from the web. As a blind screen reader user, I don't often need web-based terminals, but when I need them I need them and usually they're a11y hot garbage.
Asked in Slack where the code for ours was, pulled it down, discovered it used xterm.js which has a screenReaderMode flag which, I've learned, is essentially what VS Code's terminal uses.
15 minutes later and my PR adding this in is up for review. Waiting for it to get more testing to determine if we'll need to make it a toggle, but I fully expect this to be live within the week.
I've made a lot of small, quick win a11y changes on our platform over the past couple months. Meanwhile I met with the vendor for our collaborative wiki, suggested some very simple landmark changes that they could have made in 15 minutes, and guess whether any of those are in?
Company culture really does make all the difference, and the Fly folks have gone out of their way to empower me to change things, and to make things more accessible for me without me having to ask. A thousand points to them.
I'd appreciate help debugging why Firefox seems to rev my CPU fan heavily after either a few hours or maybe a day or so of use, sometimes locking up completely if I don't shut it down first.
Firefox shows up at the top of task manager sorted by CPU use, and exiting it causes fan use to drop.
I've disabled every single extension, including ublock origin.
Process manager shows a single entry for Firefox with something like 140% CPU use and 700+ MB of RAM, but every other entry is substantially less than 0.5% CPU. RAM use looks fine.
This is a recent generation Framework with 32 GB of RAM so I doubt it's the hardware.
GPU use is at 1 GB. Not sure if that's particularly high.
I'll get some numbers after a restart to compare and will update this thread accordingly.
Does anyone know of a CLI utility that can take a bunch of logs, parameters like which log level to focus on and maybe a few others, pipe that to an LLM, and get back an executive summary of any problems found, trends, potential issues, etc.?
As a screen reader user, I can't stand having to read logs because they're in about the least speech-friendly format ever (first the timestamp, next the level, then finally the thing I care about, rinse/repeat for each and every line.) Didn't occur to me until recently that an LLM might be a good solution for dumping in a bunch of logs and getting back some succinct analysis.
Asking because if it doesn't exist, I might score some work time to build it. But I'd rather not if someone already has.
I haven't done a #fedora install in a while. Previously I'd boot into a live environment, manually start Orca, click a desktop icon called something like "Install Fedora" and go through the installation accessibly. Now the F38 Workstation spin seems to drop me straight into the installer and I can't start Orca with either alt-f2 or the keyboard shortcut.
Is this how the installer now works? If so, is there another method to go back to how it worked before?
If it's not obvious from the above, I'm a blind screen reader user. Right now I'm pressing Enter right when the boot media loads. If there's some combo of arrows/enter that can get me to the previous live environment, that'd be very helpful.
Hey blind Android users, anyone have a favorite more advanced GPS navigation app?
In particular, I'm looking for something that will let me save my own points and navigate back to them, with both directions and bearing/distance for cases where Google Maps tries to do a dumb and take me along small access roads and such when really I just need to hang a left and walk for about 100 feet, but the app doesn't give me that info because it's too busy holding my hand.
Likely moving to a new area next week, there's a lot of cool stuff to walk to, and I'd like to just mark my building entrance so I can return after exploring things. But so many of the apps I'm finding want to hold my hand more than is necessary, and that sucks.
@objectinspace That's what speech-to-text is--converting speech to text for the model to process. Neat, wasn't sure if it was mostly a spoken interface or something else.
Took the co-op several days to reinstall my toilet and clean up the clumps of dried sewage in my shower after shitpocalypse 2 wound down because, wait for it, they apparently forgot they'd removed it and had to be reminded.
I wrote a nastygram and fed it through Chat GPT with instructions to make it less emotional and more impactful. It did a pretty decent job, but at one point I'd written something about being tired of our "fuck around, find out" approach to maintenance.
GPT translated that to me being tired of our "neglect and discover" approach to maintenance.
Neglect and discover. Love it. Punchy and vaguely ominous.
I, for the first time in 11 years, have a new physical address.
There are a lot of stories I could tell about that and the process of getting it, but holy shit I got out of practice. I used to switch addresses every couple years. I guess I forgot how awkward and anxiety-inducing the first few nights alone at a new place are. I don't know any of this background noise yet.
To be clear, I haven't heard anything alarming. It's just, places sound very different even if they're quiet, and I have 11 years of experience telling me dysfunctional home sounds like something else. Now I have to learn what maybe-hopefully-functional home sounds like.
So far so good with the new place. We'll see how I feel in a few months, but as long as we don't deteriorate to management actively gaslighting and deflecting me, it'll be a massive improvement. 😀
Oh and then there's the "I'll put my phone where I normally do, on my desk. But hey, there's no desk. Let me drop it somewhere that makes sense in the moment, do 15 things and hey, where'd I leave my hone?" Replace "phone" with half a dozen other things, and that's also my life of late. I'll figure all this shit out eventually I'm sure but right now the cognitive load is crushing.