missing stop closure after consonant clusters causing abrupt transitions. Yes yes. I know of that bug too. It showed up in Hungarian and other languages. Getting fixed. Bugs bugs bugs. All the bugs. That one is on indexing and pause frames in the new driver. Back to Python we go.
The resilience of Minnesotans is unmatched. Day and night, in snow and freezing cold, they keep showing up and standing their ground. That’s people power. That’s BlueCrew strength.💙
being able to vibe code DTrace scripts is a game changer. I'm sure the equivalent in Linux-land is going to be the same. Suddenly things you know that can be debugged but you didn't know the syntax can be solved
i've been doing so much stuff of the "i know it's possible but i'd have to read docs for 2 days" category that i just wouldn't even have started otherwise, it's such a huge productivity boost
@lain yes, this is where it unlocks so much potential. Like, I don't want to become an expert in Dtrace. I don't want to become an expert in Postgres stored procedures, or PL/pgSQL. It will take too long and I don't have a job that demands that I do it all day every day.
Life is too short to become such an expert in something that you can do by memory that you will rarely need to use anyway
@NonPlayableClown @lain it's more about knowing best practices so you don't kill your database with terrible inefficient procedures. SPs, triggers, etc. They can be immensely powerful tools that turns your database into an absolute powerhouse but also you can cause really stupid side effects
(see also: pleroma restoring from backup right now causes expensive triggers to execute on inserts)
@lain @NonPlayableClown or pleroma returning different timeline results for the exact same query/data because those functions allow nasal demons to be invoked
@i @lain @NonPlayableClown this is very likely caused by non-deterministic collation on your Pleroma database
from the mouth God (Claude):
2. ORDER BY (The Key Issue): This is where you'll see different results on repeated executions: - When values are considered equal by the collation, PostgreSQL has no stable way to order them - The sort order between "equal" values is undefined and unstable - Multiple executions may return rows in different orders - If you use LIMIT, you might get completely different result sets each time
@i @NonPlayableClown @lain we should fix this. it's a solvable problem and just requires indexes rebuilt after we switch the collation to be deterministic
Have the world's biggest headache, but must power through. Nearly all bugs are solved that crept upon me after I changed that frontend DLL. C++ still sucks, and I'm only using it because I still haven't been bothered to check out other byte-compiled (machine code, not at runtime) languages. I learned it in high school like many coders, barely used it in college only on summer breaks, never used it during employment, and yet I use it for the small link-space. So yeah. That's our relationship. It's complicated. Is all this coding making me like C++ any better? No, not really. Sure, no strict indent-checking and function definitions are easier to write than Java. It's all I'm giving C++.
Are you looking for practical ways to get the most out of using Braille and JAWS? Want tips to help you work more efficiently? Join us Thursday, January 15 at noon ET for a live webinar, “Six Ways …
Ahahaha y'all wanna know why some people hear whispering and others don't? With the latest NV Speech? I know who might. @@radiorobbe who reported it first. The Root Cause: "Dot" vs. "Comma" NVDA Sets the Locale: When you set NVDA's interface language to Spanish, Hungarian, German, Polish, etc., NVDA configures the Python process's global locale (via setlocale) to match that region. In the US/English locale, the decimal separator is a dot (e.g., 0.5, 60.0). In many European/South American locales (the "supported languages" you noticed), the decimal separator is a comma (e.g., 0,5, 60,0). The DLL Inherits the Locale: The nvspFrontend.dll is loaded into the same process space. The standard C++ string parsing functions (like atof, strtod, or std::stringstream) often respect this global locale setting. So, all the YAML files become invalidly parsed because they use a period. Ha. Wow. What a bug.
So we're going to refine nasality a bit, to actually let you specify durations rather than just having it as an "always" or "not" type toggle. Sorry y'all. allowAfterNasals will be turned into a bit more for the YAML so you can specify durations for nasalities
The weirdest part of self-hosting isn't the control, it's the silence. I logged into my dashboard tonight via the tunnel. No banner ads. No "suggested content" algorithms designed to doomscroll me. No tracking pixels firing off to a data broker in Utah, lol. Just my tools, sitting there, waiting. We really forgot what the internet feels like when it isn't trying to extract value from us every single second. It’s quiet here. I like it. #digital #minimalism #selfhosted #SmallWeb #AdFree
Feeling sad again, nothing's compiling, I broke my frontend, oh no, and not even GPT is fixing it, and I don't have pro mode anymore, and I'm just so out of options I could cry from frustration over all of it. Ugh. Stupid C++ and its cast types. I hate cast-typed languages Hate them.
Question for people relying on alt text: When one posts contains multiple images, should each alt text be written independently, or does it make more sense for one text to refer back to previous ones? Like, when the first describes shape and colour of a flower, should the second copy that description and add to it (independent alt texts), or refer back, like "the same flower [from a different angle, showing xyz]"?
So, uh right now the calculateTimes function needs to be refined to handle special cases for English vowels and diphthongs. I'm also working on that for the Frontend. Support diphthong normalization in pack files. It's not added yet but next update for sure
Jak se z nás stali geokačeři Přemýšlel jsem, jak v pochmurném prosincovém počasí dostat děti ven, až mě napadl geocaching, který se ukázal jako trefa do černého. Pár řádků o tom, jak se z nás stali geokačeři. #děti #geocachingblog.eischmann.cz/2026/01/17/j…(reakce na tento příspěvek se může zobrazit jako komentář pod článkem)
yes, yes. The output for "level" seems incomplete, possibly missing leading "l" and "və." This might be happening because the pointer iteration isn't handled properly in the script. espeak_TextToPhonemes might modify the text pointer internally, and I may need to adjust this to correctly handle wide char pointers (wchar_t*) instead of using c_void_p. Let's fix that. Why not, let's fix it all. All the bugs. Fix fix fix. No breaks.
OK, I got some questions: 1) I assume the instructions at the GitHub page to deal with phonemes are outdated, since they talk about data.py, right? 2) If I want a phoneme to sound different in Portuguese, should I create a phonemes section in pt-br.yaml and put the phoneme parameters there, or should I modify a phoneme in phonemes.yaml itself?
@clv1 nah, I updated them just recently :) the instructions should be recent. It now talks about the formats for the various pack files, so definitely do a control+shift+R just to make sure you get the latest version. Espeak is able to speak many of them, but you have to like, read them as individual letters sometimes. Yeah. I think JAWS is the only one I know where there's a way to get Eloquence to speak IPA, and supported posts about this (ruf.rice.edu/~reng/jaws-ipa.ht…) but it's tricky for sure. I will switch to Espeak sometimes in this workflow (or Speechplayer) and see how it speaks the symbol and compare as it's the only one I've seen in NVDA doing some interpretation.
So, I may not modify a phoneme in phonemes.yaml, because it would affect other languages, not only mine. Instead, I must create a new phoneme, then normalize it in my language yaml file. Am I correct so far?
@clv1 kind of. For language-specific tweaks: don’t edit a shared IPA key in phonemes.yaml unless you want every language to change. Instead: 1. “copy the closest phoneme to a new key (e.g. y_hu / ᴇ) in phonemes.yaml” 2. “in your packs/lang/<lang>.yaml, add a normalization rule that maps the IPA symbol to your new key” That keeps the phoneme table centralized, and language packs only decide which key to use.
Also, how can I hear each phoneme to perhaps discover better existing ones? Perhaps there should be a command line tool to extract and play phonemes, and test the voice each time I change a phoneme.
I have set the ESpeak dir and dll dir successfully, then opened the pack root, set my language to pt-br, and was able to play some phonemes at the list of all phonemes. But when I press the button to speak a text I wrote in the input text, it gives me an error. Pressing the button to convert a text to IPA gives an error too. The following two messages contains the errors.
for sure strange! Looks like it has to do with Espeak launching and maybe not finding espeak-data in the same path. We try to now find it more explicitly, and also if the error comes up again it should pass any Espeak errors through too. So if it appears again, it'll be more descriptive, LOL. eurpod.com/synths/NVSPPhonemeE… is the link, else the original one is the latest now too anyway.
3) Do you have special settings for NVDA to pronounce IPA symbols and unicode characters in general? Synthesisers like Eloquence, or ESpeak itself, tend to stay silent on most characters like those.
President Trump announces that from Feb. 1, there’ll be a 10% tariff on all goods from Denmark, Norway, Sweden, France, Germany, UK, Netherlands and Finland, increasing to 25% on June 1, until a deal is reached for the complete and total purchase of Greenland by the US.
I listen to someone recommend portfoliocharts.com and portfoliovisualizer.com and wonder if they're going to be accessible to me. I hope so, at least in terms of doing the things that I'd want to do, but they have "charts" and "visualizer" in their names, respectively... So we'll see. A lot of things out there weren't really designed with someone like me in mind. That's just my life. If I'm going to be using a new thing for some reason, then I'll wonder, am I going to run into any accessibility-related trouble with it?
Going to go eat something and chill a bit y'all, taking a break, let's check into the Next We Have podcast with Gareth Reynolds, love the show for chilling to. All his stuff, the more Gareth Reynolds I have in my life, the better.
It allows remapping your keyboard in many ways, and it's controlled from a simple configuration file (so, you can prepare one for your disabled friend).
You can, for example, remap the keyboard to only use its right half, if the left hand is dysfunctional (similar to the mirrored keyboard I made, github.com/clackups/qmk_firmwa… ).
Also, it should be useful for #disabledgamers , to map, for example, AWSD to different keys.
А як ви намагаєтеся спростити йому життя? Щодо мене, зараз я не вирішую ніяку проблему, але я сам сліпий, до того ж комп'ютерний ґік, люблю вселякі такі штуки, особисто коли це конфігурується у текстовому файлі.
@menelion ми випробуємо різноманітні альтернативи стандартній клавіатурі. Я зробив один фізичний прототип, але виявилося, що він не дуже зручний. Зараз розмірковуємо над новим дизайном.
А сенсорна не легше йому буде? Ну типу як у планшета, просто рухати пальцем по екрану? Взагалі, звичайно... жах. Сам терпіти не можу, коли людям мене шкода, але не можу утриматися від цього відчуття.
@menelion ні, там все складніше. Складно пояснити без відео. Рука рухається випадково, її складно націлити на клавішу. Ми уявили, що треба зменшити рух руки вздовж клавіатури, щоби простіше було націлити палець на потрібну клавішу. Тобто, кількість клавіш треба значно зменшити.
Тоді, може, щось на зразок джойстика (чи джойстиків) для вибору клавіш? Ну типу, один джойстик під одну руку, другий під іншу, і рухами вправо-вліо-вгору-вниз вибирати символ?
@BorrisInABox translation: one of our nodes has been sucked into a wormhole and will be available in approximately 9.42567 years with a latency of 7 light minutes. Your welcome.
@differentieel Yeah, that. I like the idea in principle, but just didn't want to deal with all the junk anymore. Things are worse to deal with now than they were back then.
@matthewcroughan Mostly, I got tired of getting blacklisted by <Insert_Big_Provider_Here> every couple of days, having to run around proving that I'm not a spambox or whatever.
@matthewcroughan @BorrisInABox @neil@co.uk You don't need to. You can end up on IP blacklists entirely by accident, or because $bigco just applies a weight to the entire IP range of your provider. And getting off them is hell.
@matthewcroughan @BorrisInABox It is when of them is your landlord, or your bank, or a job application, or anything else you need to interact with as an adult human.
Two of these are actual examples where I had this happen.
@Tutanota@mastodon.social Hey, I love what you do and I appreciate it, but please take the flatpak package more seriously. I hope you do this year.
Using an AppImage is not really a universal solution. For example, for those of us on musl distros (Alpine, Chimera Linux, etc.) that package is useless.
I really hope you consider treating flatpak as a first-class package because in musl it is our best and probably only option
@serrebi looks to be a timeline switching bug: it's telling me "home:" and "notifications": but not syncing it even with the setting to sync the timeline to the UI is checked. Very odd one. For sure the issue isn't that they're showing the same but that it's not shifting timelines even though the new one is announced to me. This one is tricky to pin down for sure. @Bri
@serrebi fixed: looks like I somehow had multiple instances of fastSM.exe up and running. Even simpler. Never let multiple spawn. Right now you can launch FastSM.exe as many times as you'd like. @Bri
I'm very happy with my frontend. It talks to my backend well. Saying that outloud will make me laugh every time. Sounds really dirty, but it's the truth y'all.
Indeed, I personally do notice a change in performance. To really test, just hold the tab key down. Do you hear the startings of letters consistently and does speech stop after you release it on a control right away? If so, the new frontend is not introducing performance penalties, I'd say.
Aha, how fitting for NV Speech Player. eurpod.com/synths/nvSpeechPlay… is the first version to hit the 10th file milestone for my synthesizers. It's also the first time I'm using the new directory link. This copy fixes memory corruption problems in the frontend, Post-stop aspiration block was made non-brittle now, and the applyRules allocation performance fix. No new versions for awhile until we have tweaked more languages, so I'm done with new files at least until way later in my evening or tomorrow.
one thing that I dislike about NVDA at the moment is that whenever addon updates are available, it pops up with a dialog in the middle of whatever I am doing to let me know, hey, you can update your addons! Can you just send me a toast notification? Please and thank you.
I mean you can change that. Settings>Add On Store>Automatic Updates. Honestly though it's mostly at the startup and then I don't really mind anyway, get a bunch of random shit when starting up anyway.
Sadly you can't change it to a tost notification. Like either you can disable thaht entire check I think and then you need to check yourself, or enable auto updates, which is, idk, is that really better?
First and foremost NEVER start this in a ssh connection without using screen or tmux (or equivalent). I started doing and my shitty "big tent" laptop decided to spontaneously renboot (the second time in less than 24 hours)
Kudos to btrfs-convert to have handled this gracefully.
feld
in reply to feld • • •lain
in reply to feld • • •feld
in reply to lain • • •@lain yes, this is where it unlocks so much potential. Like, I don't want to become an expert in Dtrace. I don't want to become an expert in Postgres stored procedures, or PL/pgSQL. It will take too long and I don't have a job that demands that I do it all day every day.
Life is too short to become such an expert in something that you can do by memory that you will rarely need to use anyway
NonPlayableClown
in reply to feld • • •> I don't want to become an expert in Postgres stored procedures, or PL/pgSQL.
I know a expert that works with postgres, what do you need to know for SPs?
feld
in reply to NonPlayableClown • • •@NonPlayableClown @lain it's more about knowing best practices so you don't kill your database with terrible inefficient procedures. SPs, triggers, etc. They can be immensely powerful tools that turns your database into an absolute powerhouse but also you can cause really stupid side effects
(see also: pleroma restoring from backup right now causes expensive triggers to execute on inserts)
:blank:
in reply to feld • • •feld
in reply to :blank: • • •@i @lain @NonPlayableClown this is very likely caused by non-deterministic collation on your Pleroma database
from the mouth God (Claude):
2. ORDER BY (The Key Issue):
This is where you'll see different results on repeated executions:
- When values are considered equal by the collation, PostgreSQL has no stable way to order them
- The sort order between "equal" values is undefined and unstable
- Multiple executions may return rows in different orders
- If you use LIMIT, you might get completely different result sets each time
feld
in reply to feld • • •Ludovic :Firefox: :FreeBSD:
in reply to feld • • •feld
in reply to Ludovic :Firefox: :FreeBSD: • • •@usul here you go. It works with both IPv4 and IPv6
dpaste.com/9VLZEXZ2E
feld
in reply to Ludovic :Firefox: :FreeBSD: • • •@usul example output of the script after letting it run for a minute. tail end of logs, and then the summary:
2026 Jan 18 10:20:30 20657 1001 UDP firefox 10.255.255.53:53
Command: /usr/local/lib/firefox/firefox --sm-client-id 101d024c23819c000176843403200000687830009
Local: 10.27.3.4:46950
2026 Jan 18 10:20:30 20657 1001 TCP firefox 15.204.35.21:443
Command: /usr/local/lib/firefox/firefox --sm-client-id 101d024c23819c000176843403200000687830009
Local: 10.27.3.4:26279
2026 Jan 18 10:20:30 20657 1001 UDP firefox 10.255.255.53:53
Command: /usr/local/lib/firefox/firefox --sm-client-id 101d024c23819c000176843403200000687830009
Local: 10.27.3.4:46653
^C
PID PROCESS UID PROTO COUNT
67162 unison 1001 UDP 3
95715 quasselclient 1001 TCP 8
20657 firefox 1001 UDP 12
21846 thunderbird 1001 TCP 15
2 clock 0 TCP 16
67162 unison 1001 TCP 27
20657 firefox 1001 TCP 261
0 kernel 0 TCP 4218