In addition to the huge white NS logo, there's also a tiny dark blue one, together with NMBS/SNCB logo on the new ICNG-B trains. I wonder whether it was there all the time since I haven't noticed before.
Just updated to #PeerTube v7 with the new UI. It feels really nice, fresh and modern. Good work @peertube
Also I appreciated that all my simple CSS customizations are compatible with the new UI. I hadn't have to change anything after the update. Very well designed styling. 👍
PuppyGit is a Git client for Android. Thanks to the enormous efforts by its developer who even rewrote their entire build workflow, PuppyGit was now confirmed reproducible
RSS feed: https://onj.me/media/stroongecast/index.php?xml_feed=yWelcome to episode 1 of StroongeCast, a husband and wife team consisting of Andre and Kirsten...
As of today, #StroongeCast is now available in Apple Podcasts and as of this evening because I'm stubborn and didn't want to give up until I fixed it, Spotify as well.
what was Spotify's bief? I lack the voice and tools and time and all sorts to be a regular 'caster these days, but I do like keeping up with the literature on what can and doesn't work.
@cachondo I had to include <itunes:owner> tags in things. Spotify, the absolute bloody useless tool *only* said 'waaa, sorry, email is missing.' I go looking for the correct tag which is <itunes:email>, I add it. Nothing.
I then go snooping through some other rss feeds and find out that adding itunes:name above email, and then ending the encapsulating </itunes:owner> made stuff work... Well why didn't you say so sooner you sack of specially-slow soap suds!
@cachondo I'd rso much rather stick to music, not php, but I feel damn good because I got it done. It's utterly meaningless code for a prop but I feel like I won a marathon lol
ha take the wins. I almost cried when I got my PHP certification. weeks of work, my walls literally plastered with Braille code and things, and they changed so much upgradeing from PHP 3 to 4 that it was almost a new language.
@cachondo Yeah, the old php code this thing was running on back in 2009 when Quinn basically grew the thing in his backyard is vastly, vastly different from the 2024 version. I have the old stuff to look back on, from when it was how all the TBRN show ran, and it's night and day different.
In this episode we catch you up on the things we've done in 2024, our hopes for 2025, you'll hear about Kirsten's rather strange dreams and we have a question for you, the listeners. You'll find out the answer to the question in the next episode.
Welcome to episode 16 of StroongeCast with Andre and Kirsten Louis.Good news! we're now available on both Spotify and Apple Podcasts. Just search for 'Stroon...
Every year around this time, I bring out this ridiculous clip of when I opened a champagne bottle back in 2011. It went everywhere! Sharing this recording is just tradition at this point, and I even included it at the end of my latest podcast episode for those who listen beyond the outro. Anyway, it's time for...
'Happy New Year from Andre and Kirsten Louis (AKA The puking champagne bottle)' youtu.be/SECSu6shNRk
Got time for a little story?The date: December 31.The year: 2011.The time: About 20:14.The event: I'm opening a champagne bottle, and my lovely wife Kirsten ...
In this episode we give you the answer to last episode's question, leave you with another, and then ask a series of 'This Or That' questions which you have to try and answer as quickly as possible. Questions like: 'Summer or Winter?' 'Bacon or Sausage?' Feel free to leave your answers wherever you listen to podcasts, or email them to us via the contact form.
Welcome to episode 18 of StroongeCast with Andre and Kirsten Louis.In this episode we give you the answer to last episode's question, leave you with another,...
question here though, how did the houseowner know the burgler was a burgler, did they have, "burglers are us?" written on his shirt or something? or was this person a known burgler?
@bigpawedbear Any fucker that did that, delivery person or not is going to get extremely short shrift. The obvious thought is, that person is trying to rob me. Ring the damn bell like a normal person.
@bigpawedbear Back in the 90s, we could still get milk delivered at the house. One summer evening, I forgot to close the main door, only having the storm door closed. In the morning, I guess the milk person thought this was an invitation to come into the house and put the milk on our kitchen table. I've checked all doors every night before bed since then. Lol!
@mcourcel @bigpawedbear ha. We had a dog once. Opened the door to a postman, or a parcel or something. This dog just blows right through my legs, runs upstairs and snuggles into my daughter's bed. I thought hebelonged to a friend down the road who came quite regularly on a very brief glimpse, but he'd turned out to have escaped from a garden a few streets away.
@bigpawedbear None whatsoever. I guess he just saw an open door and went for it. Very family friendly dog, obviously came from a household with children.
In this episode we discuss the fascinating topic of Synaesthesia.
Andre's version of Synaesthesia involves being able to feel certain days of the week, and also some numbers. Listen as we delve into this, and also some comments from a Mastodon thread Andre created to discover just how many others he might know that have it too.
Welcome to episode 20 of StroongeCast with Andre and Kirsten Louis.In this episode we give you the answer to last episode's question, leave you with another,...
My wife Kirsten @MoonCat and I started a podcast last September and we're having fun with it. Over 13 hours and 26 episodes in, we've covered topics like: How we first met. Never going to sleep on an argument. Double standards in a relationship. The tech from our childhood. Banning mobile/cellular phones in our children's schools and much more.
We're a blind sighted couple, I'm blind and she's fully-sighted, so we come at this from two quite different walks of life. Interested to hear more? Visit: onj.me/stroongecast and check it out for yourself. #StroongeCast
In this episode we discuss a youtube comment that we received on the previous episode, complaining that our lack of bad language made the podcast essentially unlistenable, and that 'there's no such thing as bad language' so swearing is a must. Naturally, we're here to refute this.
Welcome to episode 34 of StroongeCast.In this episode we discuss a youtube comment that we received on the previous episode, complaining that our lack of bad...
In this episode we discuss Andre's birthday and those he shares a birthday with, and hear about some of the crazy experiences Kirsten has had on busses this week and in-general. If you have an unusual public transport story of your own, please share with us. For next week's episode we're also looking for your stroonge animal encounters, be they scary, funny, completely mind-boggling or anything else in-between.
Welcome to episode 37 of StroongeCast.In this episode we discuss Andre's birthday and those he shares a birthday with, and hear about some of the crazy exper...
In this episode we discuss all sorts of Stroonge animal encounters from our own experiences, and those submitted by you. From a sneaky sneezing rat, to a late-night horse-encounter, to a squirrel squirreling itself away behind the radiator, it's all here.
Thanks to all the contributors that submitted stories for this episode.
Welcome to episode 38 of StroongeCast.In this episode we discuss all sorts of Stroonge animal encounters from our own experiences, and those submitted by you...
10, 11 years old on the train! I was indeed by myself. I'm not sure I'd let my daughter have done it. She does it now at the age of 14, but I'd feel uncomfortable letting her do it at the age i was. But yup, it was a horse. Not helpful.
My parents were quite hands-off, I was free to roam. as long as I was at school, nobody minded what I got up to in the evenings or the weekends.
@cachondo @MoonCat aah, that clarifies things. I know I did that kind of train trip alone sometimes going to Worcester, but yours sounds way more harrowing somehow.
@andrew I kept walking until I found the owner of the horses. he found it very funny that his field was close enough to the station that I could just walk a few hundred yards in the wrong direction and end up chatting to a horse.
he bought me a fish and chips and let me wait for the next train, as i recall. nice bloke.
@cachondo Good ending to the story, especially the fish and chips part. How long was the next train in? I think I would have been scared to death at that age if something like that happened to me. One good lesson though is when something like that happens in most cases as a blind person its best to try and get yourself amungst other people, don't just stand there doing nothing. I realise that is more difficult for a totally blind person but still.
@andrew trains were every 2 hours back then on the Rhymney line. Yeah, people are the best. I've rarely had a problem with the public. I see so many raging angry videos: guide dog refusals, mostly. But I've had nothing but kindness most of the time, 8 years with a dog and the rest with a cane.
In this episode we talk about the fact that it's our one-year anniversary of the podcast, how we think it's gone so far, some of the things we may cover in the future, and how Kirsten has grown in confidence over the last year. There was a time when she'd have refused to be on mic at all, but doing this weekly has made it easier for her to talk to you. We also touch on the audio quality of some of our favourite podcasts and how some of them unfortunately fall flat, which is surprising, given the budget that some have access to.
Lastly, a new, personalised birthday greeting should you want it. We're starting up a short audio segment called 'Stroonge Cards' which is a personalised greeting from your friends over at StroongeCast that sounds similar to the podcast. If you'd like such a greeting yourself, please get in-touch to tell us when your special day is, so we can record it for you. This of course doesn't go out on the podcast itself, it will be sent to you directly.
Welcome to episode 54 of StroongeCast.In this episode we talk about the fact that it's our one-year anniversary of the podcast, how we think it's gone so far...
@fireborn @MoonCat +1 to this, yes. I made sure for example that when we had our guest @ednun_p on the show, he turned on original sound so it was as good as it could be without having him send me a direct recording.
Storm Dragon pushed changes to the testing branch of the audiogame-manager project Reorganize the way notes are shown after updates so that they hopefully appear oldest to newest.
Storm Dragon pushed changes to the master branch of the linux-game-manager project Reorganize the way notes are shown after updates so that they hopefully appear oldest to newest.
Storm Dragon pushed changes to the master branch of the linux-game-manager project Updated the file name of the download. Forgot to change it from the Audio Game Manager version.
Storm Dragon pushed changes to the testing branch of the audiogame-manager project Soulblaze is now available natively in Linux, so moved to linux-game-manager.
Vimes is off to the vet for a dental, which unfortunately means: getting up at 7am, trapping a very uncooperative kitty that hasn’t eaten in 8 hours, drive to a different, farther away clinic, and then dropping said kitty off until the evening—and hoping that the sedation goes well
Vimes is fine and resting at home. The vet was in awe of how terrible his mouth was; he had to remove essentially every single tooth that was still in there—except they were all hanging by a thread and a bunch of the remaining roots had already started being reabsorbed by the surrounding tissue. This cat must have been in near-constant pain for a year, and nobody could have known.
We're still worried about the kidneys, the diabetes, and the FIV; but the mouth getting cleaned up should reduce the stress on what's left of the immune system, and stabilise everything else within acceptable parameters for a 10+ years old cat.
Day 18 - #adventOfIOSAccessibility. When building custom components, or if not relying on UIControl's attributes to configure state, it can be easy to forget to specify the right accessibility traits. These are indispensable for a good experience with VoiceOver, Switch Control, Voice Control, Full Keyboard Access...
The Internet Museum translated to Swedish becomes "internetmuseum". It is a digital, online-only, museum that collects Internet- and Web related historical information, especially focused on the Swedish angle to all of this.
Great to see you recorded for posterity. How do you feel about being part of future research for a few generations?
A similar thing happened to Digital Art a couple of decades ago with the Digital Art Museum. It spurred a few awards for excellence, most notably Vera Molnar who received the Dam Digital Art Award before her death last year: digitalartmuseum.org/history/i… ddaaward.org/
c'è una piattaforma di microblogging Fediverse che è assolutamente meravigliosa e piena di caratteristiche di design brillanti che rendono il suo utilizzo un vero piacere. Il suo nome è Sharkey, è un "soft fork" di Misskey , e merita sicuramente la vostra attenzione
Discover the beauty and joy of Sharkey, a stunning Fediverse platform that is a Misskey fork. Explore its elegant design, user-friendly onboarding, and unique features – like Antennas.
For anyone curious, I fed the code of both drivers into GPT pro and asked it to do a final comparison of how they work and are different. I mostly did this for future notes to have in case another NVDA breaks the add-on model for speech synths, I can't say this is without any possible hallucinations, but it aligns with ~95% of my observations: eurpod.com/eloq_differences.tx…
@joshknnd1982 true, but Sapi5's indexing is just... Very shit and minimal compared to what you'd get with NVDA, and some complain of a slight delay that it going through Sapi5 causes. It's for sure a valid point, though.
I've used various sapi5 tts engines for years and personally I don't notice any delay. You could just make a sapi5 bestSpeech and say: "ok there you go folks, live with it and have fun."
ha yes. But I also don't have the set of Visual Studio C compilers, and would have set up an environment with the SAPI SDK and everything. So for sure a bigger project, but if it came to the end of the road for the add-on or at least it was on the horizon, that would be the time to act on it more urgently.
Seems like Jake Gross is good at the sapi5 stuff. He may eventually make bestSpeak sapi5 who knows. I prefer sapi over NVDA addons because then it works with NVDA to balabolka to audio games but that's just me.
@joshknnd1982 yeah, you can install it alongside, but it's just a multi-gig set-up is all :) along with the overall SDK environment. Can't say it won't happen, but one step at a time I guess. The speak program is good for standalone conversion though it's true that it sadly won't work for audiogames that rely on SAPI alone.
Both Eloquence 4.7 and 4.2 received updates in the drivers: (common): Properly call DLL handle termination after switching away from either synthesizer. Previously you would fry your NVDA by switching from one to the other, as they both leave traces for one another if doing consecutive initializations. This was a problem of the past, no longer, we take care of it in both drivers for you. - Eloquence 4.7: Updated it from the 4.2 driver to also have the bgPlay function accept onDone, attempt multiple tries to feed audio, and use a lock to ensure thread-safe access to player.feed(). While this does not solve issues with Wasapi, it (slightly) improves it while keeping things more thread-safe for the older-style single callback. eurpod.com/OldEloquence2024.nv… (v 4.20) eurpod.com/Eloq47-2024.nvda-ad… Done for the night, exhausted now from reviving 2 more synths in a single day.
so it's really out of necessity if anything, most of my debug hours were spent on troubleshooting thread lock-ups and freezes. Without a second thread, if the main background thread tried to handle both the DLL’s message loop and synchronous audio feeding tasks, any delay in feeding audio chunks or flushing buffers would just make NVDA fuck off entirely or just freeze the UI up without any logs generated. I split it into eciThread for the message handling and BgThread does Audio (this approach I took from Bestspeech to solve it for the older DLLs) It may be (hopefully, from what I've seen) that the 6.1 driver lets this get managed more gracefully from a single thread since it looked more event-driven so you at least don't have things that can lock it up so easily. (Ok yeah, shit like ECIMessage.eciWaveformBuffer, eciIndexReply would have been a lifesaver if they could come from the DLL)
That might suggest that you're not calling WavePlayer.idle() when the utterance finishes. That is necessary in order to push the last buffered chunk of audio. So you need some indication of when an utterance finishes, since the next one might not come for some time.
aaaah I reverted to a prior version where I added it and it did the chunk-ends breaking, but no luck - now NVDA will silently crash with no log? TF. Think I'm just done with it for now, these are older synths and I may have to revisit it later to make them single-thread if it's possible to. But for sure a good tip on nvwave! Will keep this in mind for any other modern synth work I may get to do xD
@jcsteh your idea did help me fix eloquence 4.7 though :) I now use player = nvwave.WavePlayer(1, 11025, 16, outputDevice=config.conf["speech"]["outputDevice"]) to correctly feed it the right one, and the BGPlay thread does call player.idle() which is good. Making it run on one single thread though caused that silent crash oddly, and that's with logging set to "debug." so I'm just like SMH
Err, sorry,I'm actually wrong anyway, which is hilarious because I wrote that code. idle() isn't necessary to push the final chunk of audio. But if you don't call idle(), you won't get callbacks (onDone) for the final chunk until another chunk is fed.
not sure if this is something I need to fix on my end, or this might be a bug, but I noticed if you try taking the 4.2 one up higher than 60% on the speed, it crashes. no error or anything like that, but just stops talking. I have to use the ctrl+alt+n key to force shutdown, and restart NVDA. as long as I don't go any faster than 60% on that one I am fine. now the 4.7 one I would use at 70 percent, but too fast for me, had to slow that one down. the previous version I had to take it up to 70 percent, to make it fast enough for me, but this particular version, the latest one as of right now I had to slow down as all of a sudden 70 percent was a bit too fast for me. I hope this helps.
I am currently using the 4.2 one for right now, but I just can't set it faster than 60% without it crashing which is just a bit too slow for me, but I love it! this is classic tts's at there finest.
@JamminJerry very odd as I have it fully max at 100. could be another one of those hard ones to ever get pointed down since not everyone's seeing it. Unless someone else jumps in to work or fix these drivers I most likely won't as they are just too old of software synths, I only cared to get them functional under modern NVDA
ok, I don't know if this matters or helps you in any way, but I have put the inflection back to 30 percent, and now I can take the 4.2 version to at least 71 percent.
@KaraLG84 strange! works here, but I'm not familiar with that bug! Wasn't really a jfw 3.2 user when it was around. As long as the dll is in the eloquence folder it should load though, inside synth drivers in the add-on's folder of it inside eloquence folder there are the 2 dlls it loads.
All the DLLs are there. What happened was that when your pc automatically switched daylight saving off, JAWS crashed. And it was a bug with this version of 'Eloquence, I think. @MutedTrampet would no.
@KaraLG84 It's really strange! For me I tested it out by setting my date to 03-09-2025, and now the driver, well, it silently crashes without even a log entry. Wow. That's nuts. I hope I can do something that @MutedTrampet or others can recommend, maybe we gotta fool the dll to think if it's a different time, check if system condition is DST and apply a fix? That's my guess anyway but good find! Weird quirk.
@KaraLG84 Perhaps try adding or subtracting one hour (3600 seconds) from the time you pass to the license function during DST. I remember trying this but not sure which one worked.
this fixed it I think! Going to patch the driver with this change in the license request function: ``` # Patch for old DST bug: if localtime indicates DST, shift time by 1 hour # If adding doesn't help, try subtracting instead (lic_time -= 3600). lic_time = int(time.time()) if time.localtime().tm_isdst > 0: lic_time += 3600 lic_time ^= 0x39AB43F2
eci.eciRequestLicense(lic_time) h = eci.eciNew() for i in (rate, pitch, vlm, fluctuation, hsz, rgh, bth): vparams[i] = eci.eciGetVoiceParam(h, 0, i) return eci, h
``` Will be uploaded in a moment. In my testing, moving my clock to Mar 09 at 1:59:59 and then clock changes, I can still load the DLL. Test it in a few minutes (update: uploaded) and let me know if you still encounter issues. Huge thanks for this idea.
@KaraLG84 Also curious, why are you calling the old Eloquence driver Eloquence 4.2? The version number in the resources of ECI32D.DLL is 3.3, and the language DLLs are at 3.0.1.
ah yeah, originally I was looking at the wrong DLL's version :) Great point! Easy to change on the site. Can't believe it's 3.3! Makes me wonder if there's an intermediate version now even more between 3.3 and 4.7 as that's a huge jump :D (fixed version in add-on manifest as well)
was there ever a 1.0 or 2.0? Wonder if we would have recordings of it somewhere. I wish there was a way to pinpoint down when the voice so drastically changed, clearly there must have been other versions developed between those that slowly got it there. The more you unravel about that product the more fascinating history it has :D
3.3 also appears to accept the current community dictionaries, although don't think these can be language specific like newer Eloquence. I renamed them to main.dic and root.dic from the community dictionaries, loaded them and surprisingly that worked. I don't see a way to disable abbreviations as this appears to be built-in to the language and I couldn't find a function in 3.3 that toggles them off or a speech parameter, would be nice to add that as a setting if it's ever found.
@TomGrant91 @KaraLG84 From what I remember, dictionary loading for multiple languages wasn't implemented in that code, and the release notes for Eloquence 5.0 claimed that all languages shared a single dictionary set in older versions. Recordings of 1996-era versions of Eloquence can be found in this article. archive.nytimes.com/www.nytime…
@MutedTrampet @TomGrant91 @KaraLG84 this is huge help! The other issue I had was with pp1 tags, it would announce a back quote after parenthesis symbols with it sent. I don't think Pp2 and PP3 did this so I may be able to add some form of pause shortening.
There probably were many intermediate versions between that which aren't publicly available, or at least not in the known sources. ETI was known to produce lots of minor versions of major releases, example: Eloquence 5.0. That version was officially released on February 8, 2000 according to the release notes, with the earliest build I can find being version 5.0.000.0004, compiled on March 31, 2000, and the last known builds were of version 5.0.3.8, compiled on December 10, 2001.
@MutedTrampet @KaraLG84 aaah I love your eloq research. It's been super helpful and I added the abbreviation setting into Eloq 3.3 driver. I'll see what I can do about the other PP tags for the shorten pauses, it may be that I add the setting as a choice slider like the IBM drivers and people can just choose then in both 3.3 and 4.7. Even if the tags add some broken punctuation for 3.3, at least folks can decide.
Kanál s přednáškami z konference OpenAlt, která se koná tradičně v listopadu v Brně a zaměřuje se na otevřený software, data, IT bezpečnost, DIY a IoT.
@stepan @fredbrooker Nahraval jsem je primo do PeerTube (vcera vecer 2023, dnes rano 2024), ale porad jeste bezi transcoding, takze se na kanalu neukazuji vsechny. V playlistech ale jsou.
Alright, here we are. Old eloquence 4.2 driver for NVDA 2023-2024: eurpod.com/OldEloquence2024.nv… (works both with Wasapi on or off) Eloquence 4.7 driver for NVDA 2023-2024: eurpod.com/Eloq47-2024.nvda-ad… (Fixed to work fast with Wasapi off as well)
@twynn I tried that version but it wouldn't index right for me :( would just read the first half of a dialog and then, boom, stop. So when it was re-done it was very hastefully done so, maybe just ran through a py2 to py3 parser but it meant that a lot of newer stuff wasn't in it.
Thanks for all your efforts over the past... five or so days in reviving these old synths. BTW, Eloquence 4.2 doesn't allow you to switch to other languages besides American English, but it sure sounds far different to what we have today that's for sure.
@twynn added these! I just only put ENU initially when debugging it for the language dictionary list. We have two french, two spanish, UK, US, and Italian in that version now.
@twynn As a very much non-eloquence-using person, what's the difference between all these versions and IBM as well? They all sound similar to me. Didn't grow up with them so just don't really get it.
A Coast Guard Commander Miscarried. She Nearly Died After Being Denied Care. —
U.S. service members have long faced strict limits on abortions, even when used to resolve miscarriages. Under federal #law, the #military will only pay for abortions in cases of rape, incest or to save the mother’s life.
U.S. service members have long faced strict limits on abortions, even when used to resolve miscarriages. Under federal law, the military will only pay for abortions in cases of rape, incest or to save the mother’s life.
A Further Review Update on the Choyong LC90 By John Figliozzi, NASWA Journal Shortwave Center Editor Prepared December 12, 2024 Having previously reviewed the overall performance of the Choyong LC9…
Removed three dead domains (catgirl.life, vtuberfan.social, natehiggers.online). They have no DNS records, or redirect to a subdomain with no DNS records. Will properly comb through for dead domains at some point later.
You'll want map_err if you have function A that returns a Result with one type of error, and you need to call it in function B which is required to return a Result with a different type of error. I did serious Rust programming for like 2.5 years before I recall having to use map_err. It'll be there if you ever need it, but it's not necessarily something you have to learn right away to get productive.
While you're changing tactics, though, I'd suggest thinking about starting a real project in Rust rather than working through exercises. I never actually worked all the way through the official Rust book myself. I didn't seriously make an effort to learn Rust until I had a real project I wanted to do in it.
Eloquence 4.7 driver for NVDA: Used in older JAWS versions (35, 3.7): eurpod.com/Eloq47-2024.nvda-ad…. Modernized indexing : On end-of-utterance (0xFFFF), flush once, then send all indexes and None, and set speaking = False. Similar to BeSTspeech logic, the code stores intermediate indexes (currentIndexes) during the utterance and only sends them all at the end. For me, it was interesting to see how older and newer Eloquence drivers were different: The newer V6.X DLLs schedule tasks like flushing audio, sending indexes, or finalizing utterances via a background thread. Older Eloquence did a lot more just from the Callback thread alone, huh. also: Instead of sending indexes as soon as they appear, the newer ECI stores intermediate indexes for the entire duration of the utterance. Only after the utterance fully completes does it send all intermediate indexes, followed by a "None" signal to mark the end. This prevents premature indexing events and ensures that the indexing data sent to NVDA accurately reflects the final audio and why I was stuck for 3 hours debugging it.
@sclower I can kind of see why though, it doesn't have the random pops and clicking noises that eloquence 6.1 at faster rates has, that's a nice part of that one.
I'm noticing some unusual artifacts with 4.7 which I don't recall hearing before. Specifically, words that end in a long e sound odd, and the combination of r and a long e, like the word "tree," are slightly garbled. Aside from that, it's working well.
@sclower wow yeah you're right, that's an odd one. Wonder if that's a change we ended up with after playing through Nvwave or not, hmm. Only really there if you're looking for it, but once you notice it you can't un-notice it either from knowing it's there, ha.
@sclower I think I remember it happening in JAWS 3.3 comedies and a demo I got some time back. Words like tree, three and bree, for example, do sound garbled, that's normal as far as I know.
@TomGrant91 No, it's definitely not normal, at least with the version of ELoquence that shipped with 3.5. Does the add-on use the newer wave speed compression algorithm?
@sclower @TomGrant91 hmm, is that a driver setting? Not that I'm aware of. I even swapped in the files from eloq-4.7-2023 and it still had that odd artifact! Even though 2023 4.7 did not, it's super strange. I did get to fix the character-by-character reading issue though.
@sclower @TomGrant91 also verified we're using same bit depth / sample rate and set up the engine with the parameters from the older 2023 driver. A headscratcher for sure.
@TomGrant91 Yeah. It sounds like the original audio is slower and being externally time-compressed in some way. But this is all observation; I don't know the internals of that driver.
@sclower @TomGrant91 what I'm curious about: Why is Eloquence 5 technically the 6.1 dll, and where are the dll versions between 4.7 to 6.1. Personally I'd love to get the pacmate-Style eloquence working, which I believe was closer to 4.7's driver model than 6.X, but might be thinking of it in reverse.
@TomGrant91 Really? Based on the audio I've heard from the Pacmate, I always assumed they were using 6.1. That would have been about the right time for it.
unless they had set the inflection to be super high, but it's also the same driver that would have shipped in JFW 4.1 and 4.2 (then 4.5 got the flatter, more monotone 6.1 we know today) which... gives me an idea... Oh no.
It's so weird hearing this again. Though I did use Jaws back then, it was almost exclusively with a Doubletalk PC. I mostly avoided the use of Eloquence for everything until just a few years ago, except for Talks, when that was still a thing.
I just found out that, if you want the rate to go faster, for the languages in eloq.cfg, for example, 1.0/enu, you can change the version from 2.0 to 4.7, and then in eloquence.py you can change MaxRate to around 200 or 300, and the synth will go much much faster after that.
Wait, aren't you doing it wrong, then? Indices are supposed to be sent to NVDA as soon as the synth fires them, they should only be premature in the case of a few samples ahead due to buffers. IBMTTS is actually doing it wrong and it means that the phonetic punctuation add-on does not function. So don't take your pointers from it.
@x0 Thanks for pointing that out. I really appreciate the detailed insight into how indices should be handled. I’m still learning how to integrate everything correctly, and updated both drivers with what you mentioned: Removed storing indexes in currentIndexes for both drivers. Instead, we now Call onIndexReached(lp) immediately when we get an intermediate index (lp != 0xFFFF) in the callback. Also, the final utterance index (lp == 0xFFFF) still triggers a flush and then a finalization, but doesn’t involve delayed indexes anymore. It meant I could Simplify finalizeUtterance() since we’re not handling a queued list of indexes so it cleaned up the older driver quite a bit. Not sure if people will notice any user-facing differences, but nevertheless this is great info to have.
Glad I could help, that's my understanding anyway but pretty sure it's correct, that's why indices exist, to give immediate, real-time updates of exactly where the synth is. NVDA itself actually injects index marks into the stream to help with synchronization, say between sentences so it can move the cursor in say all. Phonetic punctuation uses them to decide where it should inject sounds into the stream, asking the synth to pause at the same time. So that has to be precise, and buffering, as IBMTTS does, throws that off.
@x0 oh that's fascinating. I think I was used to the older simpler drivers where we had to generate our own index (even if not exact) so it's interesting to me that even the older eloquence drivers were sophisticated enough to do this.
Well they had to be, as eloquence actually has a callback option to communicate visemes. That is the most precise you can possibly get, it's sending you a note of the mouth shape corresponding to every phoneme its putting out in real-time for speaker synchronization.
@x0 DO you know if Eloquence Threshold does it right? I use that one rather than IBM TTS for performance reasons, I always found it performed better and now i'm wondering if this is the reason why.
@nick @x0 it's yet another packaged eloquence. IBMTTS never gave you the binaries so had a feeling of legality about it. Threshold was just installable and worked. github.com/pumper42nickel/eloq…
Eloquence synthesizer NVDA add-on compatible with threshold versions of NVDA (2019.3 and later). Supports Python 3 and new NVDA speech framework. - pumper42nickel/eloquence_threshold
@cachondo @nick @x0 ooh ooh, but they fail to remove the DLL handle when the synth is terminated in Threshold, bad bad. Loool. Only if you have both installed really, IBMTTS does correctly unload it so you can go from that to "eloquence" but not the other way around.
@x0 @matt @cachondo @nick I've actually always wanted to hear those! Never got a chance to play with them. I tried the 6.7 version with the 22K sampling rate and it had a weird high-pitched artifacting to it. Now that I've used eloq at 16K and 22K going back to 11K is like someone switching on an AM radio xD
That's because the concatenative voice binds to voice 4, Rocko, rather than voice 1. Also, you got the formant synthesis to operate at 16 and 22 kHz with NVDA? how? What DLLs are you using?
@x0 @matt @cachondo @nick nah, sadly just 22. Ideally I'd use the 16K version but that one I think is only found on Android and iOS (when choosing high quality / sampling rate). 22K is not bad but sounds a bit forced, although v6.6 at least removes that high-pitch squeal in the upper parts of speech.
@matt @cachondo @nick I never had the 6.4 or 6.6 DLLs, they run the formants at 22kHz? Where can I get those? IBM's 6.7 doesn't support it, it just ends up running at double speed.
@matt @x0 @cachondo @nick interestingly the DLL copyrights of 6.6 still say 2002, unless they were too lazy to change it from 6.1 values. It may be that several versions of these were concurrently developed (certainly the weird reversals in stability might indicate this)
you have to run the command: C:\Users\User\AppData\Roaming\nvda\addons\Eloquence\synthDrivers\eloquence>regsvr32 ibmcmm.dll (change it to your username and folder where you store the files for it)
@matt Version 6.6 of IBMTTS came out in February 2003. IBM continued maintaining IBMTTS for many years after SpeechWorks stopped developing Eloquence in 2002. Their last big version update was 7.0, released some time around 2005, and the last recompiled IBMTTS language DLLs I can find are from May 2007.
wow really there was a 7.0? Wonder how that one performed, and why 6.1 somehow became that defacto standard in all AT products. Maybe it just wasn't quite as stable even in 7.0
@MutedTrampet I can only speak for Serotek, but as far as I know, we didn't try to license IBM TTS, only Eloquence, because the latter was what everybody wanted. And when we finally got access to Eloquence from Nuance in 2009, the latest available was 6.1.
@matt That was indeed the last version ever developed for PCs, as far as I can tell. Kurzweil got 6.1.0.2 in 2006, but that only applied to a few files, and was probably meant to primarily resolve the conflict that occurs if Eloquence and IBMTTS are running at the same time, which IBM was aware of but never got around to fixing.
7.0 was really obscure, I can't even find a changelog for it and I also had to rip it out of various installers. Its main features seem to be the introduction of version 3.2 of the Torrent engine which was used by the 22 kHz concatenative voices, and for US English, a bunch of DLLs that seem to be some kind of heavily rewritten synth engine with individual DLLs for different components, with names like enuparser.dll, enuphonetizer.dll and enuprosody.dll, which are oddly never used by enu50.syn and I can't seem to find how exactly they were meant to be used. Version 7.0 actually completely fixed the stability issues of 6.7 with the June 2005 build of ibmeci.dll, but it never made it to the assistive technology market for some reason and 6.7 ended up being the last update that would ship in AT products, even though that version is extremely unstable.
@x0 @matt @cachondo @nick Do these need any specific driver to run with NVDA? I can't seem to get ibmeci66 to load if I change the DLL name in the settings, and the eci.dll that comes in that zip file doesn't seem to support 22k. Am I doing this all wrong?
@guilevi @x0 @matt @cachondo I could load it undeer the latest IBMTTS driver but you have to register the ibmcmm.dll file first with regsvr32 and then the ibmeci66.dll file will load and then you gotta go to voice settings > sample rate to change it in NVDA
@guilevi @matt @cachondo I think I already have a version of that registered for 6.7, do I have to register this one too? Ibmcmm.dll is for the concatenative voices.
@x0 @guilevi @matt @cachondo yeah not sure if they're compatible, I've had conflicts before where it would weirdly error out when I had the wrong one registered from a viavoice install
@x0 @matt @cachondo NVDA still dies after registering that DLL when trying to load ibmeci66. I don't recall installing any particularly weird Eloquence/IBM versions (or any versions at all) on this machine so not sure where I went wrong. This is exactly what I remember from every time I tried to get anything nonstandard about IBM to work before hahaha
@matt @cachondo @nick Though that one goes by the name viavoice, which is IBM's version of that engine, and its rules are different, including its internal dictionaries. Mohammed's repo actually has a special dictionary set for it because of some words it fixed and others that it broke. Jake's site has it as a giant 7z file called viavoice something or other.
@TomGrant91 @x0 @matt @cachondo @nick wow I wonder how the performance on one of those voices would be and if they're even that natural sounding. My guess though if I haven't heard a lot of folks use it it must have been pretty trash for day to day usage.
@nick That may have been mitigated by now, there were some changes to threshold itself that changed buffer intervals to try and make it a bit less floody, and then WASAPI happened. I've got an open issue on IBMTTS to integrate this behavior now that it's feasible, they're still using index buffering. THey don't buffer until the end of the stream, they buffer by larger chunks to avoid short feeds. This is likely the performance difference you're noticing, as I think threshold is sending the chunks immediately, whereas IBMTTS actually sends a chunk after receiving the next one, unless it's less than the normal chunk size, then it sends them both at once. The old NVWave really didn't like being sent a tiny chunk at the end.
@nick Threshold is here. Also take note of how it handles pauses, in ECI, pauses, the `p annotation usually triggered by the break command, are dependent on rate! At least in 6.1 it is, impirical testing would be needed for the others. github.com/pumper42nickel/eloq…
Eloquence synthesizer NVDA add-on compatible with threshold versions of NVDA (2019.3 and later). Supports Python 3 and new NVDA speech framework. - pumper42nickel/eloquence_threshold
@nick The reason pauses need to scale correctly is because of phonetic punctuation, which knows the length, in milliseconds, of the sound it needs to play.
@x0 @nick it really is snappier for sure, wow. It's probably what I was noticing on the older drivers, they've gotten a slight improvement in just how quickly they respond to commands I feel like.
Salesforce says it will hire 2,000 people to sell AI software, a month after saying it would hire 1,000 for the task and two years after laying off 7,000 staff (Jordan Novet/CNBC)
SCOTUS rejected ISPs' challenge to a NY law requiring ISPs to offer $15 or $20/mo. services to low-income customers, by refusing to revisit a lower court ruling (Jon Brodkin/Ars Technica)
Rebuilding much of my infrastructure in NixOS again because I'm sick of shit breaking and being hard to fix. In the process I've been revisiting services I run to ask whether they're worth maintaining or whether a simpler alternative would do.
Started poking Nextcloud today and oh wow is its app ecosystem extensive. I thought it was just a nerdy contact/calendar/file-syncing app but no, you can really build out your own household-specific portal with it.
Right now for instance I'm playing with the Cookbook app. So far I've pasted in a few recipe URLS and have gotten back nicely-accessible, no-fluff copies of those recipes, saved locally, linking back to the originals, with spinners that seem to scale them by serving size. That's really neat.
So I don't think I'll be replacing it with Radicale quite yet. It can apparently somehow parse emails and extract travel itinerary data into calendar events with kitinerary but I'm not sure how. That's one more proprietary service down.
Right now each service like Gitea and Nextcloud gets its own native NixOS container. Nix config assumes things like one instance per server for many of those. Fortunately it's pretty easy to spin up a Nix config in a container vs. directly on the host, and Docker/OCI containers are pretty easy to declaratively build and run in your engine of choice if Nix containers/services don't work.
🚧Disclaimer: This blog post has been edited and rewritten using Claude and Gemini.
It's always good to be on a lookout for ways to improve your day-to-day workflows and be able to automate away at least some work that isn't very creative o…
I like this snippet from the article: "Just like the rest of our image and video generation models, Veo 2 outputs include an invisible SynthID watermark that helps identify them as AI-generated, helping reduce the chances of misinformation and misattribution." blog.google/technology/google-…
TIL about Crabeater seals, an animal that looks like every other seal out there, like the recipe was fifty percent chonk, fifty percent derp, moisten and add fippers, but then when it opens it's mouth you find out that it is secretly hiding the answer to the question, what if my teeth had teeth.
🚨help ginny pay for necessities—she lost her rented house and many things to flooding and theft. she needs continued funds for a place to stay, and care for herself and her animals. the stress (and the impact on her health) has put her in and out of the hospital.🚨
she is an amazing person who has suffered greatly. she is disabled, an orphan, and was targeted by the Proud Boys.
she survives on community support, so please support her 💖
NEW! When you support my work for $15/month or more, a glossy postcard with personal note and featuring photography from my current photo/poetry project (Come Outside) will be yours.
Tennessee Ernie Ford is a legend and I hope I've done his classic tune "Sixteen Tons" justice in this re-imagining of it. Thanks for watching! LINKS AND SUC...
Martin Hoffmann
in reply to Ondřej Caletka • • •