I'm just going to say, Wow. That's a pretty incredible result. I had no idea you could clone a voice with such a short snippet of audio these days.
I've had this dream of talking to the Enterprise-D computer since I was 8 years old. Midlife-crisis me still has that dream, but hooked up to Home Assistant so it can actually do useful things too. A couple months back, I went looking around for "clean" samples of Majel's voice as the computer but didn't have a lot of luck. Even though there are three television series and several movies, pretty much all of them have some amount of background noise, bleeps, bloops, or warp core thrum. (As this one does.) There may be modern ways to clean those up without affecting her voice much, but I haven't dug into that yet.
There are a few audiobooks narrated by Majel Barrett but obviously her role as the computer was proper voice acting and so the books would not be good source material.
There were also a few games/CD-ROM (Omnipedia) with some samples, but they did not bother to post-process them for that lofi Enterprise-D computer feel. Can _probably_ be replicated fairly faithfully after the fact, but I only know a _little_ about audio post-processing.
According to her son (Rod Roddenberry), Majel sat down in a studio and recorded audio samples specifically for the purpose of having her voice cloned someday for future Star Trek stories. However, those haven't been released publicly. (And likely never will, but I can dream, can't I?)
Edit: I played with your samples and the time it takes to generate the output audio is pretty brutal. Too slow for interactive use. Maybe that's a limitation of the HF-hosted app, though.
So I guess there's "masking" and "adapting." TFA doesn't seem to differentiate between the two.
I don't "mask" around people I'm comfortable with and know me well (friends and family), but I definitely "adapt" in situations where people don't know me well and I am eager to fit in and/or match their energy.
Occasionally you do have to "adapt" even with friends and family, depending on the situation. I like to crack jokes pretty much all the time, but if I did that when someone was describing something serious (a close relative dying, for example), that would quickly ruin a relationship.
Even neurotypical people do this, we put on different personas around different groups. Neighbors, co-workers, close friends, acquaintances, parents, kids. That's normal and healthy because all of those relationships are different. That doesn't mean your core personality is wildly different, just that your engagement with each of those groups is slightly different because the expectations are different.
I'm open to trying it, but my gosh, having one 'w' offset slightly by a few pixels from the one right above it feels like it would drive me bonkers eventually.
And: doesn't this result in text that "jumps around" as you type?
Many otherwise open-source chat apps are "open-core," they tie certain features to a subscription. Can be things like chat history, voice calls, video calls, but a very popular one is SSO and AD/LDAP integration.
I just can't get super upset about this. Sure, OTC companies are duping customers with marketing, but what's new about that? As the person holding the money, it's my job to look at what is effective and what the active ingredients are in any given product. Or ask my doctor/nurse/pharmacist what to do, if I can't be bothered to make the effort myself.
When I want to get irrationally angry about something in a department store, I'll walk over to the shampoos, which for some reason always have a whole entire aisle dedicated to a single product, when they all do literally the same exact thing, just with different scents and advertising budgets baked into the sticker price.
There is a guy on YouTube who did very thorough and well-done presentations on airliner crashes and mishaps and one of the reasons they were so good was that he was a very experienced pilot himself. He was able to give deep insight into the technical details, the industry, and the challenges that pilots face. He always talked at length about those in the context of the incident he was covering, which was how his videos were so much more interesting compared to your typical "accident documentaries" thrown together by outside amateurs which are frankly the majority of videos in this space on YouTube.
But since the last year or so, I can't watch him anymore. He sold his channel (and his brand, literally himself) to some kind of YouTube content company and the videos he puts out now are just not watchable. From what I can tell, he mainly does only the presentation now, with only a minor amount of editorializing. Other people seem to do everything else. The visualizations are impressive but the video title/thumbnails are pure click-bait (to the point of being factually false), the videos are WAY longer than they need to be, and he'll repeat the SAME information multiple times just to stretch the time out to 45 minutes to an hour.
I like a good story, but it's really hard to pad out most disaster videos into that amount of time unless you have something more to offer than say, "well, the official crash investigation said this and that." His videos now feel a lot more like those old Discovery Channel documentaries that were basically surface-level filler content in between the ads.
Every company I have worked for in the last 15 years or so has required me to sit through mandatory training in which we are specifically told never to offer any government a bribe in exchange for the ability to conduct normal business there.
It's pretty that amazing OpenAI doesn't have such a requirement.
No, that's about right in my (very limited) recent experience. It's _very_ easy to spend as much at a fast food restaurant these days as you would at a sit-down restaurant, especially if you don't do one of their combo meals, or add one or two extra items to your order
The crazy high prices and general unhealthiness aside, my main beef (if you'll pardon the pun) with fast food places is that more and more of them are taking orders via AI and/or requiring you to download and install their app to place an order.
I've never corrupted an SQLite database _file_, but an FTS5 index is remarkably easy to mess up if you don't read and understand the docs to the letter and which parts of them apply to your exact use case. When you do a thing wrong with FTS5, you either get a generic ambiguous error, or silent corruption of the index. It's not a black-box extension to you can just enable with no understanding of how it works under the hood, I learned that the hard way.
There are several variants of Qwen 3.6, the MoE models are performant on Strix Halo, but the 27B dense model (the one spoken about in TFA, and generally regarded as the best of the group in terms of quality) is not so performant: https://kyuz0.github.io/amd-strix-halo-toolboxes/
No, even MoE models need to fit into (V)RAM. MoE has faster inference because only a subset of layers are used to predict the next token, but the set of layers used changes with every token.
Quantization is a trade-off, though. The quality, while still perhaps good enough for many tasks, is not as good as the full 16-bit weights that the model was designed for/released with.
> it's very easy to add a TFTP server and configure everything to serve whatever you want.
In your own homelab or in a small company, sure.
But the nice thing about proxyDHCP is that in a larger company, if the network engineering team hands you a subnet to play in that has DHCP forwarding configured in the router already, and you want to do PXE in it, you can just deploy your own proxyDHCP server without any extra red tape.
Or in my case, I just don't like to have configuration for a single service scattered around my network devices if I can avoid it.
Crazy theory: Maybe Codex watched too many YouTube videos. I have noticed on YouTube that many younger narrators say, "I quickly did X," when "I did X" would have been more just as well and usually more truthful. Why did you have to _quickly_ cut that piece of wood when the video clearly shows cutting it at a perfectly normal speed?
Also, please try not to anthropomorphize LLMs, they hate it when you do that.
charles dot hn at bityard dot net