fastfinge

For those of us using screen readers, this is a way bigger deal. Honestly I probably shouldn't use a bluetooth headset and a bluetooth keyboard for my banking. We focus so much on SSL/HTTPS and wifi security, but I wonder how much effort goes into wireless keyboard security? Not nearly as much, I'd bet.

Problem was that I usually only discovered the issue when I went to read the book lol

I never did that, my connection was too slow to want to take up someone's DCC slot for like a day to get an entire movie. Remember all the frustrating idiots who would share .lit files, but forget to remove the DRM from them?

Ah, good to know. Back in my day, when we had to walk a hundred miles to school in the snow, up hill both ways, IRC was the only place to get ebooks. I'm guessing it's just the old users clinging on now.

Man, I’m getting flashbacks to my days running omenserve on undernet. I had no idea people were still doing this! How does the content compare to places like Anna’s archive these days?

Prophecy approved companion is excellent! It gave me all the feels. It’s both extremely funny, and extremely poignant as the main character learns who she is, what’s really going on, and her intended roll in it all. It’s one of the few series where the reader knows exactly what’s happening from the start, but the fact the main character being slow to catch on isn’t frustrating.

Also, if you don't feel comfortable building bookworm from source yourself, and you feel like you can trust me, Here's a build of the latest bookworm code from github for 64-bit Windows: https://www.sendspace.com/pro/dl/rd388d

If you use Bookworm and use the built-in support for espeak, you can get up to 600 words per minute or so. Dectalk can go well over 900 words per minute. As far as I know, cocoa tops out at around 500 words per minute. So all of the options accept piper should be fine for you.

No, Mistral 7B can't describe or work with images. Thanks for answering!

It really depends on your use case. If you want something that sounds pretty okay, and is decently fast, Piper fits the bill. However, this is just a command line TTS system; you'll need to build all the supporting infrastructure if you want it to read audiobooks. https://github.com/rhasspy/piper

An extension for the free and open source NVDA screen reader to use piper lives here: https://github.com/mush42/piper-nvda

If you want something that can run in realtime, though sounds somewhat robotic, you want dectalk. This repo comes with libraries and dlls, as well as several sample applications. Note, however, that the licensing status of this code is...uh...dubious to say the least. Dectalk was abandonware for years, and the developer leaked the sourcecode on a mailing list in the 2000's. However, ownership of the code was recently re-established, and Dectalk is now a commercial product once again. But the new owners haven't come after the repo yet: https://github.com/dectalk/dectalk

If you want a robotic but realtime voice that's fully FOSS with known licensing status, you want espeak-ng: https://github.com/espeak-ng/espeak-ng

If you want a fully fledged software application to read things to you, but don't need a screen reader and don't want to build scripts yourself, you want bookworm: https://github.com/blindpandas/bookworm

Note, however, that you should build bookworm from source. While the author accepts pull requests, because of his circumstances, he's no longer able to build new releases: https://github.com/blindpandas/bookworm/discussions/224

If you are okay with using closed-source freeware, Balabolka is another way to go to get a full text to speech reader: https://www.cross-plus-a.com/balabolka.htm

Can Mistral describe images yet? Not sure if it's multi-modal or not. If it could that would be a super useful feature for those of us over on rblind.com. And/or is the code available somewhere for us to hack in something like openrouter and spin up a copy?

Personally I find myself renting GPU and running Goliath 120b. Smaller models could do what I’m doing if I spent more time optimizing my prompts. But every day I’m doing different tasks, and Goliath 120b will just handle whatever I throw at it, no matter how sloppy I am. I’ve also been playing with LLAVA and Hermes vision models to describe images to me. However, when I really need alt-text for an image I can’t see, I still find myself resorting to GPT4; the open source options just aren’t as accurate or detailed.