mastodon - Link to source

I'm watching people in my feed screaming at each other over Firefox's "AI kill switch" this morning with some trepidation.

As far as I'm concerned, Firefox already has an AI kill switch. It's called browser.ml.chat.enabled, I set it to false more or less the day it appeared, it hasn't mysteriously popped back on since, despite angry posts to the contrary, and that's been that for me. It's disabled every "AI" feature I find objectionable. I'd prefer if Mozilla leadership would sync up with reality on occasion and stop deciding to put this paid placement trash into Firefox in the first place, but at least there's a reliable way to get rid of it.

Everything else people have been screaming at Mozilla about? I'm not sure I see the problem. The little model you can download to summarise web pages for you? I wouldn't trust it, and so I don't think it's necessarily a productive use of Firefox devs' time, but at least it's opt-in. The other little model you can download to help organise your tab groups? I don't use it much, but this one seems more practical, and it's also opt-in, despite the occasional angry report of it slowing down people's browsers even without having been downloaded. Liek, bro, maybe try closing a Slack tab or two.

I don't want ChatGPT in my browser, or Claude, or any kind of world burning data centre LLM pretending to be our new AI god. browser.ml.chat.enabled = false does that for me. It would be even better if it wasn't there in the first place. But tiny, focused ML models doing nominally useful things? I not only do not see the problem there, I'd like some of them to be part of the Web platform rather than just the browser. I've been wishing since they launched them that Mozilla would make an API available to web sites out of those translation models of theirs, for instance.

I'm monitoring the situation, as European heads of state like to say, but so far, despite the posturings of their C-suite types, it doesn't seem like any critical Mozilla resources are being diverted away from maintaining the Web platform into AI boosterism. Every new Firefox changelog is delivering on what it should be delivering on, and it's only occasionally that I see a new "AI" feature advertised. Compare that to a product like VSCode, which has been completely consumed by the cancer of slop production with only one in a hundred changelog entries being about building an actual damned code editor, and I'm not feeling all that alarmed about Firefox just yet.

I know this is Mastodon, but sometimes I just wish people would entertain having opinions that can have some nuance in between "burn the world down so the AGI can live" and "Butlerian Jihad now," you know?

mastodon - Link to source
Stranger Things complete. Hard to believe a show that I've watched since middle or high school ends when I'm an undergraduate. That was sort of the only thing I watched on Netflix. But the service is disappointing now. There are barely any good shows or movies to watch. And if they are, they're are average at best, or they got canceled for reasons which usually involve money and whatnot.

mastodon - Link to source
Isfugl - Alcedo atthis - Eurasian kingfisher
#fugle #fotografi #birds #birding #birdingphotography #herlevfotoklub #nature #wildlife #2730foto #photography #dof #dof #dofbirdlife
#Isfugl #Alcedo_atthis #Eurasian_kingfisher
#photography #wildlife #nature #birds #birding #fugle #fotografi #birdingphotography #herlevfotoklub #2730foto #dof #dofbirdlife #isfugl #alcedo_atthis #eurasian_kingfisher
in reply to Tom Grant

mastodon - Link to source

Tamas G

@TomGrant91 oh, if you place the .exe where your soft voice add-on lives or the DLL files do, it'll auto-detect it. But because we can't technically bundle it all together in the repository, it's possible that someone builds the speak program on its own and would need to then manually give it where that file lives. I doubt the folks behind SoftVoice are around anymore, but I had to stay cautious. Even though the Eurpod page is public, it's not tied to the wrapper project so if anyone reaches out and says, "take this down, you're redistributing our speech engine!" I can just yank the file off Eurpod and not touch my repo.
@Tom Grant

mastodon - Link to source

RE: mas.to/@gabrielesvelto/1149428…

With regards to Grok generating CSAM, I'd like to point out that all the image generation models are probably trained with CSAM in the first place. The kind of scraping that was done to train them is indiscriminate and the companies that make them don't care, so god knows what's inside the training data. My recommendation still stands: stay tf away from them

Gabriele Svelto (@gabrielesvelto@mas.to)

Also, torrenting porn at that scale means that Meta's models are pretty much likely to contain NCII and even CSAM. Keep the fuck away from them.
Gabriele Svelto (mas.to)
This entry was edited (6 days ago)

reshared this

mastodon - Link to source
So. Because GPT uses Jupyter internally, when it reviews files, they get truncated for it. Poor thing. It spends half its time trying to figure out why the code is truncated when in reality it's Jupyter and the output truncating it. Ha. Can't believe that.
Gemini is good, but it can only do code changes inline. GPT will actually patch your file for you if you are specific enough and upload it. But then of course it spends half its time trying to figure out whether a piece of code exists or not because it got truncated. Unreal.

mastodon - Link to source

PSA We are happy about the wealth of privacy focused classic email providers (proton, tutanota, posteo, mailbox.org etc) but they are not geared for instant secure messaging. Do yourself, family and friends a favor and use delta with any of the many chatmail.at/relays

Or setup a relay of your own if you feel like it.

Or use a a dedicated chat email address on your own or any classic email server that has reliable service and no obnoxious rate limits or annoying spam handling.

Cheers.


Chatmail: Relays

Chatmail provides FOSS infrastructure for interoperable, secure, speedy and reliable end-to-end encrypted messaging. Check out clients as Arcane Chat, Bots or Delta Chat today!
chatmail.at

ArcaneChat reshared this.

in reply to Attilax

mastodon - Link to source

Delta Chat

for setting up contact with a friend, see delta.chat/en/help#howtoe2ee -- there also is a mode that doesn't require a camera/scan-qr code but sharing an invite link.

Delta Chat: FAQ

What is Delta Chat? Delta Chat is a reliable, decentralized and secure instant messaging app, available for mobile and desktop platforms. Instant creation of private chat profiles with secure and i...
delta.chat
in reply to Delta Chat

hubzilla - Link to source

Fedir Ustynov

@Delta Chat
Have you tried calling the person and attempting to onboard them using only voice?

The only thing that came to mind was something like: "Take a sheet of graph paper and start shading in squares. Shade the first square. Now the second. Now the third..." Of course, dictating a link out loud is not much easier.

I hope that onboarding two or three friends a day for about a week will make it clear that all the options you suggested are not great. :)

@Delta Chat

mastodon - Link to source
We could for example attach an Eloquence audio sample, then ask for a synth that sounds similar. In case the AI couldn't make it from scratch, we could ask whether another synth could be the basis, for example ESpeak's klatt variants. @fastfinge @jscholes @cachondo @FreakyFwoof @amir @ZBennoui @pixelate @Tamasg
@James Scholes @Tamas G @Sean Randall @Andre Louis @Zach Bennoui @Devin Prater :blind: @🇨🇦Samuel Proulx🇨🇦 @Amir
in reply to 🇨🇦Samuel Proulx🇨🇦

mastodon - Link to source

Tamas G

Sort of my thought sadly. It's gotten better, no doubt, you can now get AI to spit out 60 KB of slop in one go, wow progress. xD So context improved, maybe a slightly better skillset, but the amount of time you'd spend debugging and seeing which step it went wrong on, especially for all the low-level plumbing an engine needs is brutal. @clv1 @jscholes @cachondo @FreakyFwoof @amir @ZBennoui @pixelate
@James Scholes @Sean Randall @Andre Louis @Zach Bennoui @Devin Prater :blind: @Amir @Cleverson
in reply to Cleverson

mastodon - Link to source

Tamas G

so it's a lot.
1) Signal generation (the “voice box”)
This is the DSP engine: glottal source → filter(s) → radiation → output audio.
2) Control model (turning phonemes into trajectories)
You need to decide how parameters move over time:
•
How /a/ differs from /i/ in F1/F2
•
How consonants inject noise and shape transitions
•
Coarticulation: the “smearing” of neighboring sounds into each other
•
Rules for duration and transitions (and exceptions)
This is where “it works” becomes “it sounds like a person instead of a kazoo.”
AI helps, but you still need a design. AI can implement whichever model you pick (Klatt-style rules, gestural targets, diphones-with-formants, etc.).
3) Text to phonemes (G2P)
For English you can ship a dictionary + rules.
•
normalization (numbers, dates, abbreviations)
•
tokenization
•
stress rules
•
phoneme mapping5) Voice design + tuning
Even with a perfect engine, it’s easy to end up with “robotic but intelligible” rather than “pleasant.”
This is typically the biggest time because it’s:
•
parameter tables
•
hundreds of little exceptions
•
endless listening tests
•
DSP engine: days to a couple weeks
•
G2P + normalization: weeks
•
coarticulation + durations: weeks to months
•
prosody: weeks to months
•
tuning to ‘nice’: open-ended
@fastfinge @jscholes @cachondo @FreakyFwoof @amir @ZBennoui @pixelate
@James Scholes @Sean Randall @Andre Louis @Zach Bennoui @Devin Prater :blind: @🇨🇦Samuel Proulx🇨🇦 @Amir
in reply to Cleverson

mastodon - Link to source

Tamas G

yeah, I think if a team came together for it, splitting that work perhaps by person or 1 to 2 people per section, could really work. I know I could be useful here at the later shaping stages, so do count me in, it's that architecture creation and initial rules I'm a bit out on. But yeah, not against on being included.
@fastfinge @jscholes @cachondo @FreakyFwoof @amir @ZBennoui @pixelate
@James Scholes @Sean Randall @Andre Louis @Zach Bennoui @Devin Prater :blind: @🇨🇦Samuel Proulx🇨🇦 @Amir
This entry was edited (2 days ago)
in reply to Cleverson

iceshrimp.net - Link to source

🇨🇦Samuel Proulx🇨🇦

And also UX researchers, probably. I can't articulate why eloquence is better than dectalk, for me. Neither, I bet, could Andre articulate what makes Orpheus better than Eloquence, for him. So to get something that makes the largest number of people as happy as possible is a classic UX research problem, probably involving massive surveys, rating and ranking of samples, and so on. I work with the kind of people qualified to do this, and it's a unique skill-set in and of itself.
in reply to muž s klapkami na očích

glitchsoc - Link to source

Štěpán Škorpil

I heard something.
It seems to me like community hysteria and bad company response on the topic.🤷 Well, I know a little about this, but what I heard, they are paying to multiple open source projects, right? And one of them is Omarchy/Hyperland? Which is run by a open far right idiot? That itself is not making them facist.
This entry was edited (6 days ago)

mastodon - Link to source
There was a bug (thanks to someone for discovering, I won't name names) where your own custom settings in SoftVoice applied to all variants, rather than them using their own parameters upon speaking. This is fixed partially: You can now choose a different voice but if you adjust pitch again from what you set, it'll apply. This second part is more tricky, as the engine doesn't tell us, "Hey, this is the pitch I set it to when the voice changed." Instead, you call one of the variants and the engine resets with the parameters it has internally for them. Tricky tricky. Nonetheless, it's (mostly) solved, what a relief.
Re-Download 2025.3.5: eurpod.com/softvoice-2025.nvda…
This entry was edited (6 days ago)

mastodon - Link to source
I have created a Speak Window for SoftVoice that implements the wrapper. Besides allowing you to choose variants, it also lets you save a wave file of the spoken text. Now you can proudly feature SoftVoice in your audio productions. Enjoy. eurpod.com/softvoice-say.zip - this is also for @jaybird110127 who asked for this one.
Note: a bug exists where variants like Martian do not apply all of their voice values, actively in progress to solve.
@Jayson Smith
This entry was edited (6 days ago)

reshared this

in reply to x0

mastodon - Link to source

Tamas G

@x0 @TomGrant91 so if one of those callbacks can receive marker events, then maybe. Right now the problem is our order of things:
• we call SVTTS/SVNarrate,
• we intercept waveOut,
• we detect end-of-utterance,
• NVDA sends the next chunk,
• the engine inserts its own little “tail” silence and/or waveOut open/close gaps happen. And that's the pause. Because I can't get a callback from SoftVoice saying "I'm done processing this chunk now, send me the next" it's all very sequential and the wrapper is the only thing that can tell NVDA a DoneSpeaking event properly occured
@x0 @Tom Grant
in reply to x0

mastodon - Link to source

Tamas G

sigh. After fiddling with this DLL for the latter part of the day, it's just buggy. Even if you use it in the old driver that directly called it and used MME with no hooking. It will stop before finishing longer chunks, so it'll sound as though it's stopped before speaking has finished. Very unfortunate. No amount of hooking can fix it as the same symptom appears in the wrapper. Tried to remove the trim silences logic to fix it, thinking it was the cause. But nope. So it just never sends the last few samples either. What a shame.
This entry was edited (5 days ago)
in reply to x0

mastodon - Link to source

Tamas G

darn. I think I know why: It's a trial DLL. It looks in the registry under: SOFTWARE\SoftVoice\ and checks. If it's not there, it bails out on a timer. There's strings for "Evaluation period has expired (demo versions only)
" too. So it would need that timer cracked by someone who's good at, well, cracking. :D (which will definitely not be me xD )
This entry was edited (5 days ago)
in reply to Tamas G

mastodon - Link to source

Andre Louis

PW Web Speak was one of the first ways I accessed the internet back in the day, my school got a 'Super-fast' bonded ISDN line at 128 Kbps. You could only access it in the library and on one machine but wow, the internet felt so magical back then, even if it was via a weird browser.
Still, that was my introduction to Softvoice.
@Tamasg @x0 @jaybird110127
@Jayson Smith @x0 @Tamas G
in reply to Jayson Smith

mastodon - Link to source

Jayson Smith

Wow this thing is buggy! The following text crashes the thing! I'm just guessing about the inflection command but here goes: {{rate 90; pitch 100}}Never gonna give you up,
{{pitch 50}}Never gonna let you down,
{{rate 50; pitch 200}}Never gonna run around and dessert you.
{{rate 75; pitch 100}}Never gonna make you cry,
{{pitch 100; rate 150}}
Never gonna say goodbye,
{{pitch 50; rate 50; inflection 0}}Never gonna tell a lie and hurt you.
in reply to Seedy!

mastodon - Link to source

Tamas G

@cubic @TomGrant91 yeh it's not reinitializing it from the default male variant params.
I have updated ApplyWrapperSettings to be more aggressive.
1. Force-set the personality twice on the very first run.
2. Add a small Sleep after setting the personality to let the engine's internal table update.
3. Explicitly Reset the engine state when the app starts.
eurpod.com/softvoice-say.zip
@Tom Grant @Seedy!

mastodon - Link to source
To je tak, když 4 dny v týdnu strávíte zavření minimálně 8 hodin v plechové bedně s 80 lidmi, kdy půlka z nich chrchlá jak před smrtí a kapesník, roušku nebo aspoň si zakrýt hubu zvládne tak 10%.
Ta interní cirkulace vzduchu tomu taky moc nepomáhá.
A pak se mě okolí diví, že jsem nesociální.
Než jsem začal řídit autobusy, byl jsem nemocný naposledy v 19 letech. Teď každý rok minimálně dvakrát.
in reply to The Matrix.org Foundation

mastodon - Link to source

chebra

@matrix should, but often aren't. I'm saying as a long time daily user of matrix. The number of problems caused by this is so high that I totally understand how a non-technical person will end up with a broken account sooner or later, and never come back. Dismissing feedback won't help. Prioritize compatibility. Prioritize stability. Prioritize "users will never see Unable to decrypt". New features don't matter if it keeps breaking.
@The Matrix.org Foundation
in reply to chebra

mastodon - Link to source

The Matrix.org Foundation

@chebra not trying to dismiss feedback, but we’re not aware of any problems caused by matrix.org running something incompatible with other clients or servers. what are you actually thinking of? unable to decrypt errors are not to do with matrix.org instance compat (and should be largely gone since sept 2024 as per youtube.com/watch?v=FHzh2Y7BAB…)

Unable to decrypt this message - Kegan Dougal

Day 2 - LAB4
YouTube
@chebra

mastodon - Link to source

My (public) weekly emails celebrate three years today. The fact that almost 500 subscribers willingly want my nerdy details every week fascinates me. I love you. 💚

daniel.haxx.se/blog/2023/01/10…


My weekly report on email

Starting this week, you can subscribe to my weekly report and receive it as an email. This is the brief weekly summary of my past week that I have been writing and making available for over a year already.
daniel.haxx.se
This entry was edited (6 days ago)
in reply to Seedy!

mastodon - Link to source

Tamas G

@cubic nah, they're definitely the same variant/ voices. It's a bit more static when I go back to the original crappier softvoice driver, but setting pitch to 4 on the new one makes the voice have the same tone and tambor. So not sure. Those variants are internally stored. Of course it's easy to make any voice clip if you manually mess with the NVDA exposed controls, they are mostly there for fun or customizing a variant - but some settings don't match with the variant you chose so then it can't apply properly. So, not much else to do, this is probably as good as it'll get, which is to say, better than the prior NVDA driver.
@Seedy!
in reply to Seedy!

mastodon - Link to source

Tamas G

@cubic oof. well I made a 2025.3.6 version of the driver at htttps://eurpod.com/softvoice-2025.nvda-addon. It's weird because for me, robot and martian sound actually robot and monotone.
Quick “is it fixed?” test you can run in NVDA
1. Set pitch to something obvious (ex: 9) while on Male.
2. Switch to a custom personality (Martian/robotic/etc).
3. Speak a sentence.
It should not keep the high pitch because we reapply that variant's own pitch there.
@Seedy!

gotosocial - Link to source

En Leonieke, wat doe je dan eigenlijk met al die borduren die je de afgelopen jaren hebt gemaakt?

Eh, ja, die zitten (al dan niet netjes) opgerold in diverse theedoeken.
Bij sommige borduren heb ik een ontvanger in mijn hoofd, maar bij anderen? Ik vond ze leuk om te maken, maar na een paar jaar is het wel duidelijk dat ik er verder niets mee doe.

Dus ... in de aanbieding, gratis te ontvangen, incl. verzending binnen Nederland, de volgende borduren (niet allemaal gesigneerd/gedateerd, want ja, beetje lui). :embroiviki:

1/n

#LeoniekeBorduurt #borduren #PointDeCroix #CrossStitch #embroidery #FiberArts #LTA26

#embroidery #crossstitch #fiberarts #leoniekeborduurt #borduren #pointdecroix #lta26
in reply to daughter of lilith

mastodon - Link to source

Sylvia

@marta Android's lifecycle management basically recreates the entire activity ("app screen") when a device gets rotated because the "screen size changes".

It's 100% on me. It's much easier to prevent now, but it's very old legacy code so needs a rewrite. However as solo dev of a hobby project I have to pick priorities and make sure it stays fun so I can keep going so I just haven't gotten to it.

Part of me just hopes someone else comes along and fixes this code for me so I don't have to 🙈

@daughter of lilith

mastodon - Link to source

In the USA, in most states, you can get a restricted driver's license at age 16.
Anyone who is born today, and is qualified to get such a license, is three years younger than the original iPhone.

Quite frankly, it scares me that there will be people driving who were not even a thought of a thought when the first iPhone was announced, possibly with older siblings who also were not around at the time the iPhone was released, who might be in their first or second year of college by now, but here we are.

This entry was edited (6 days ago)
in reply to Jeffrey D. Stark

mastodon - Link to source

Tamas G

@jstark Sometimes though change happens because people did work on the inside. When Microsoft decided to add MSAA (Microsoft Active accessibility) to Windows, it wasn't because there was a lawsuit, but we had people like Kelly Ford and others working inside to make change happen. Wouldn't call that one fear-based, although perhaps pressures by orgs did help that come along, since people were so concerned about transitioning from DOS to Windows. But I definitely know that that one wasn't lawsuit, but in 1997, primarily due to pressure from disability advocates and government agencies. @yatil
@Eric Eggert @Jeffrey D. Stark
in reply to Tuta

mastodon - Link to source

Tuta

Hi there, new accounts can get disabled if they are marked as possible mass signups and need manual approval. As a private email service, we do not ask for phone numbers or any other personal information so we must be very strict with this which can lead to the fact that real users get locked out. If this is the case, please contact our support at hello@tutao.de, and we'll be happy to look into this and approve the account.
in reply to Tuta

mastodon - Link to source

egg_shapped_egg

oh, my bad. I thought that fediverse accounts, and content was accessible across multiple instances 🤷

Your reply is fair and has been mentioned by the OP too. It also is the point of the discussion

I have a paid account with you so I feel like it won't be an issue for me. It's just sad if new users are discouraged from using Tutanota based off one users experience who got flagged as spam

⇧