I’ve observed some notable improvements when benchmarking with the CachyOS kernel on NixOS via Chaotic’s Nyx using moderately old hardware:

https://programming.dev/post/38304031

Haven’t yet tried replicating the same comparison on newer hardware, but would be interested to see what others have tested. Any observations?

Does anyone know of an Android app to install an additional 3rd party TTS engine that can then be configured to point to a custom Open-AI/Fast-API endpoint for self hosting higher quality voices that are not easily run/fit on mobile hardware?

Some poignant questions for these new platform requirements:

• How do you anticipate this being used against journalists and advocacy groups?
• What research and statistical quantification will be done to evaluate the amount of harm these restrictions can inflict?
• What precautions or safeguards will users have against malicious state actors or capitulating corporations?
• How can developers protect themselves from liable damages due to service interruptions caused by third party verification?
• Do you foresee legal restrictions in rollout due to national security concerns from differing nation states?

I’ve really enjoyed using Kokoro for generating audiobooks:

• https://github.com/hexgrad/kokoro
• https://huggingface.co/spaces/webml-community/kokoro-webgpu

Be sure to first try using this convenient API wrapper:

• https://github.com/remsky/Kokoro-FastAPI

Note that not all the modelled voices in Kokoro-82M are of equal quality, given disparities in limited training data from reference speakers. However, what’s cool is that you can prescribe polynomial weights to multiple voices tags, enabling you to synthesize different variants weighted more heavily from the highest quality voices.

• https://huggingface.co/hexgrad/Kokoro-82M
• https://huggingface.co/hexgrad/Kokoro-82M/blob/main/VOICES.md#american-english

One current limitation for Kokoro is that there’s no way to prescribe emotion or intonation procedurally using markup tags like SSML in the source text, unlike other models like Orpheus. But Orpheus sometimes generate weird hallucinations like repeating sentences, injecting new phrases, appending radio silence or filter words, and generally increasing the tempo of words per minute as a sentence progresses. Still, this may be of interest if you want to add emotion like fear or urgency to your generated dispatches, and manage to tune the input temperature you want for the model.

• https://github.com/canopyai/Orpheus-TTS
• https://github.com/hexgrad/kokoro/issues/36

However, Kokoro is a lot more compute efficient and audibly consistent, requiring less scrutiny or manual supervision. The author behind Kokoro now also looks to be working towards an emotional variant as well:

• https://huggingface.co/posts/hexgrad/182254197987426

Reference project I’ve been following for audiobook generation:

• https://github.com/prakharsr/audiobook-creator

I think it stands for Managed Service Providers.

I’m still using an old UC Gateway, doubt my homelab will outgrow it.

One thing I appreciate about NixOS is the ability to use overlays and override package sources. For example, overlays can be used to selectively install unstable and stable packages alongside each other:

• https://discourse.nixos.org/t/mixing-stable-and-unstable-packages-on-flake-based-nixos-system/50351

While there may be caveats, this approach has been working for me just fine, as I can install VSCode from unstable to get the most recent monthly releases as they roll out, but then pin the rest of my desktop environment to stable to limit anything else shifting underneath me unexpectedly.

Looks like they introduce the use Traefik with NixOS here:

• https://youtube.com/watch?v=_e6K0dTTjwI

How does Traefik compare to a reverse proxy like Caddy?

If there was a simple Debian based distro that I could declaratively manage via a single config file, I think I’d try it. I.e. not using Puppet or Chef that can only bootstrap a system state, but something to truly manage a system’s entire life cycle, including removing packages and anything littering the system file tree. But since there isn’t, I’m using NixOS instead.

Having a DSL to declare my entire system install, that I can revision control like any other software project, has been convenient for self documenting my setup and changes/fixes over time. Modularizing that config has been great for managing multiple host machines synchronously, so both my laptop and desktop feel the same without extra admin work.

Nixpkgs also bolsters a lot of bleeding edge releases for the majority of FOSS packages I use, which I’m still getting used to. And because of how the packaging works, it’s also trivial to config the packages to build from customer sources or with custom features. E.g. enabling load monitoring for Nvidia GPUs from btop that many distros don’t ship by default.

I’m not the original author, even with the YouTube title being as is, but what do you mean? Perhaps relying that the desired services exist as nix packages, or that nix packages have desired defaults or exposes desired config parameters?

There are two other nix media server config projects I can think of, but I think this approach mostly facilitates the install, but not the entire initial config setup, given that a lot of the stack’s internal state is captured in databases rather than text config files. So simplifying the backup and restoration of such databases seems the next best thing to persist your stack configs with nix.

• https://github.com/rasmus-kirk/nixarr/issues/3
• https://github.com/notthebee/nix-config/blob/892878506094fa2189aac0ae4e9d3aea3d0a2fbc/homelab/services/backup/default.nix#L58

On top of that, it’d be nice for the Bluetooth spec to roll out a higher bitrate version of HFP, as it’s common 16 kHz monaural configuration is awful when listening to multimedia while on video calls, like for remote watch parties or just listening to music or playing video games while hanging out on discord. I ended up just buying a USB to TRRS adapter with pass through Power Delivery in order to use my Android device with proper AV quality.

• https://en.wikipedia.org/wiki/List_of_Bluetooth_profiles#Hands-Free_Profile_(HFP)

I hope compatibility with git submodules gets ironed out soon. I’d really like to have multiple branches of a superproject checked out at once to make it simpler to compare source trees and file structures.

I’m using a recent 42" LG OLED TV as a large affordable PC monitor in order to support 4K@120Hz+HDR@10bit, which is great for gaming or content creation that can appreciate the screen real estate. Anything in the proper PC Monitor market similarly sized or even slightly smaller costs way more per screen area and feature parity.

Unfortunately such TVs rarely include anything other than HDMI for digital video input, regardless of the growing trend connecting gaming PCs in the living room, like with fiber optic HDMI cables. I actually went with a GPU with more than one HDMI output so I could display to both TVs in the house simultaneously.

Also, having an API as well as a remote to control my monitor is kind of nice. Enough folks are using LG TVs as monitors for this midsize range that there even open source projects to entirely mimic conventional display behaviors:

• https://github.com/JPersson77/LGTVCompanion
• https://github.com/cali-95/LGTVLinuxCompanion

I also kind of like using the TV as simple KVMs with less cables. For example with audio, I can independently control volume and mux output to either speakers or multiple Bluetooth devices from the TV, without having fiddle around with repairing Bluetooth peripherals to each PC or gaming console. That’s particularly nice when swapping from playing games on the PC to watching movies on a Chromecast with a friend over two pairs of headphones, while still keeping the house quite for the family. That kind of KVM functionality and connectivity is still kind of a premium feature for modest priced PC monitors. Of course others find their own use cases for hacking the TV remote APIs:

• https://youtu.be/io_AFseYleE
• https://youtu.be/76wAEJOdBq8

Hello HN! OP, you’ve reached the front page of Hacker News:

• https://news.ycombinator.com/item?id=36565842

That looks neet. Although I suspect this would succumb to the same cross post discoverability issues where URLs pointing to the same video would not match string for string. A better approach might be to facilitate inline embedding of HTML video players into Lemmy using browser extensions, where user scripts could be used to preview youtube links or re-write them to nocookie, allowing the Lemmy web UI to still avoid the use of cross-origin scripts by default.

• https://programming.dev/post/135777

Found the full transcription for the video from OP author:

• https://lemmy.world/post/647059

Note to self: use youtube.com instead of youtu.be
for better cross post detection and lemmy integration

For programming tutorials, yep, I also prefer reading documentation instead. Although, it looks like this tutorial these folks put out doesn’t have much of anything you could copy from, like terminal commands, given its a recorded walkthrough in using the graphical web UI. YouTube also now allows for searching the auto or manual transcription text, which is handy when creators always forget to include timestamped chapters.

Checking the issues tracker for RES, there’s not yet any mention of lemmy or kbin:

• https://github.com/honestbleeps/Reddit-Enhancement-Suite/issues?q=is%3Aissue+is%3Aopen+lemmy

Perhaps you could ask there. I’d also recommend checking out the Lemmy Plugins & Userscripts community:

• https://sh.itjust.works/c/plugins

For posterity, I later stumbled upon the authors original post here:

• Tamper/Greasemonkey script to reformat interface to a style recalling old.reddit
  • https://sh.itjust.works/post/53395

The community that this was posted from also looks interesting:

• Lemmy Plugins & Userscripts
  • https://sh.itjust.works/c/plugins

Check out this issue:

• https://github.com/soundjester/lemmy_monkey/issues/7