The State of Modern AI Text To Speech Systems for Screen Reader Users: The past year has seen an explosion in new text to speech engines based on neural networks, large language models, and machine learning. But has any of this advancement offered anything to those using screen readers? stuff.interfree.ca/2026/01/05/ai-tts-for-screenreaders.html#ai#tts#llm#accessibility#a11y#screenreaders
I've started using eSpeak-ng. It's grating, but I can crank the speed up way higher than any other TTS I've ever used, especially the fancy AI shit that simulates breath draws and lip movement
@PepperTheVixen The reason it's grating is because unlike Eloquence and dectalk, Espeak only uses formant synthesis for the vowel sounds. For consonants and plosives, it instead uses concatenative recordings based on human speech. That's why even when you switch to a voice that sounds less sharp, the "t", "b", "p", and other sounds are still too sharp. This seems to be the primary cause of the fatigue most people experience while using ESpeak.
There is a 32-bit compatibility layer in the works for NVDA itself (although it currently only references SAPI4). But with any luck the need for every add-on to implement its own will go away.
Link to issue number:
Summary of the issue:
64 bit NVDA can no longer support SAPI4.
Description of user facing changes:
SAPI4 is again availble on 64 bit NVDA.
Description of developer facing cha...
@jscholes@cachondo@FreakyFwoof@amir My understanding is that when this comes to addons, it's going to require some kind of secure addons API/layer. And it won't be ready for 2026.1, or maybe not even 2026.2.
I see the "Secure add-on runtime" on the roadmap, with the note that "The first version of the runtime will provide support for speech synthesis and braille devices."
I don't see any implication that any 32-bit compatibility layer will only work for secure add-ons, which is hopefully a bit of a leap.
Still, the fact that people don't know what will or won't be happening, or whether their preferred synthesiser(s) will work or not, continues to be a big part of the problem. @cachondo @FreakyFwoof @amir
@jscholes@cachondo@FreakyFwoof@amir That's my assumption because the only things that really need a 32-bit compatibility layer are speech synthesizers and braille devices.
Regarding ESpeak-ng, AFAIC, the main complaint from users is its base tone, which cannot be solved by simply making new variants. In this regard, how about improving its MBrola voices?
@clv1@jscholes@cachondo@FreakyFwoof@amir The issue is that both of these are effectively concatenative or parametric, rather than formant, systems. So they will never be as intelligible as eloquence.
@FreakyFwoof@cachondo@amir You should be able to get either Gemini or Codex to help you, depending on what AI you have access to. The workflow would be: 1. download gemini-cli or codex-cli, and get them installed and configured. 2. clone all of the sourcecode from github.com/fastfinge/eloquence_64/3. Delete the tts.txt and tts.pdf files, so you don't confuse it with incorrect documentation. 4. Find any API documentation for orphius that's available, and add it into the folder. 4. Run codex-cli or gemini-cli, and tell it something like: "Using the information about how to develop NVDA addons you can find in agents.md, and the information about the Orphius API I've provided in the file Orphius-documentation-filename.txt, I would like you to modify the code in this folder to work with Orpheus instead of eloquence."
It will go away for five or ten minutes, ask you for permission to read and write the files it's interested in, and then give you something that mostly works. Now, build the addon, run it, and tell it about the errors and problems you have and ask it to fix them. In the case of errors, include the error right from the NVDA log, and for bugs and problems, tell it exactly what it's doing wrong, and exactly what you want it to do instead. Keep doing this until you wind up with a working addon.
Think of AI as a particularly stupid programmer, and you're the manager in charge of the project. You should be able to get this done without paying anyone.
Well there's already a 32-bit addon for Orpheus floating about. I'd still rather pay someone competent to do it, even if they use AI. Proper programming terms would help narrow down the broken bits. I'm just an audio guy.
@FreakyFwoof@cachondo@amir Yeah, you can get AI to modify the 32-bit addon for you. That's how I got the first two eloquence prototypes; it helped me understand the problem and what approaches would work and what wouldn't. If you give it the 32-bit orphius addon, and the 64-bit eloquence addon, it should be able to understand the working approach to make an addon 64-bit, and make the modifications itself. The reason to give it the 64-bit eloquence addon as an example is so it doesn't decide to go down the GRPC route and include protobuf and a bunch of other nonsense.
Oh happy days 😊 that was the voice that used to come with the Hal ScreenReader isn't it? That was my first ScreenReader after my accident back in 1996 and I seem to remember the plug-in synth was called something like Apollo two or thereabouts such happy memories 🙂 but not really I used to sit up till about 4 am banging my head against the brick wall trying to figure it out but hey ho
@luiscarlosgonzalez@cachondo@FreakyFwoof@amir I didn't try Kokoro, because it cannot achieve a real time factor of 1 on CPU. By that I mean, to be fit for consideration with a screen reader, a text to speech voice must be able to generate one second of speech in one second or faster. In general, Kokoro takes two seconds to generate one second of speech. So it's not suitable.
Supertonic works way better on my 2012 machine than Kitn. That one just cuts off things all the time and just doesn't sound good at all. I'm pretty sure it's not supposed to do that. Very weird.
Really interesting article. I'm particularly passionate about this subject, I've been fascinated with TTS for a number of years. I've trained many voices, both for Piper and some of the newer LLM based systems, and while I can't speak to the speed issue, training data is extremely important.
What you feed into these models has a big impact on the voice's performance overall. If you give it stuff scrape from the web, random audiobooks that weren't optimized for TTS, things like that, you're not going to get good results for the type of work screen reader users do every day. This applies to all of these systems, not even just neural networks. The latency / responsiveness issue is something we'll have to solve at some point, because I don't think using TTS systems last updated in 2003 is going to work out in the longterm, as much as I love Eloquence.
In my ideal world, we would have either a machine learning based or formant system that is easy to train / maintain. Big companies have lost interest in on device TTS, not even just for screen reader users. Many of the solutions being put out now are cloud based, and while developers are still creating on device models, as said in the article, they're not optimized for our needs and may never be. I think we have to take matters into our own hands and figure this out, but I believe with enough people we can make it happen.
@ZBennoui Agreed. I think blast bay is close to the right track. If only it was open and the issues pronouncing words were fixed. The speed and sound of the voices are top notch.
@ZBennoui We need a good formant system. Machine learning is useful for setting the model parameters. But I think the word to phoneme rules can’t be a neural network, because they have to be reproducible and modifiable. Even here though, machine learning could help though. I’d love a system where a user could submit a recording of a word, and the system could create the phonetic representation.
Yeah I completely agree, I happen to know Philip and have been talking with him extensively about his experiments with TTS. I can't go into a ton of detail, but I'll say what he said publicly. The system he's using is a hybrid approach of neural networks and formant synthesis, where he trains a model to output formant frequencies based on the audio data he feeds into it. I won't pretend to understand all the details, this is way above my pay grade, but as far as I understand this has never been done before by another developer.
@ZBennoui Yup. I just wish he wasn’t also trying to train his own phonemizer, because I really believe that has to be reproducible and modifiable for users. I’ve swapped multiple emails with him about an NVDA addon. But he’s pretty set on sapi for now until things stabilize both on the NVDA side and on his side.
@amir I too think we can make it happen by taking matters into our hands. I don't know how to code, but I'm at disposal to work on Portuguese language support, e.g. improving pronounciation rules, when time comes.
@FreakyFwoof The major problem with incorporating AI TTS into screen readers is latency. Maybe you can use it for say all, but it is not suitable for navigation. There are a couple of tiny TTS with low latency like Piper TTS, but the quality is not the best. Also multi lingual support and pronunciation for many uncommon words are issues for AI tts.
@jaybird110127 Yup. That and the fact that the sourcecode isn't getting updated; just getting it to keep compiling is a huge effort. There is a 64-bit build, but it doesn't actually work. I consider dectalk pretty much dead, even though the source is available.
isn't it possible to "pregenerate" the speech with all the necessary IDs so that you can navigate and interrupt at will? Just as one generates SSML from rich text (including maths formulas) before generating speech.
It would even be better to catch intonations, breaths and others, unchanged instead of letting the TTS generating a "pleasant full phrase" (a wrong expectation).
I find your post intriguingly close to the emerging reaction against the Ai-generated #mundaneslop.
@polx Maybe, but probably not. Doing that would result in a lot of wasted resources generating text I'm never going to listen to. Think about the average user interface: dozens of menus, and toolbars, and ads, and comments, and so on. Plus, the text changes constantly, on even simple websites. That's not even taking into account websites that just scroll constantly. It might be possible to create some kind of algorithm to predict the most likely text I'll want next, but now we've just added another AI on top of the first AI.
I think a better solution might be to make the text to speech system run on different hardware from the computer itself. This is, in fact, how text to speech was done in the past, before computers had multi-channel soundcards. This has a few advantages. First, even if the computer itself is busy, the speech never crashes or falls behind. Second, if the computer crashes, it could be possible to actually read out the last error encountered. Third, specialized devices could be perhaps more power and CPU efficient.
The reason text to speech systems became software, instead of hardware, is largely because of cost. It's much cheaper to just download and install a program than it is to purchase another device. Also, it means you don't have to carry around another dongle and plug it into the computer.
Hello fedicrowd! Does anyone have the #Fairphone 6 with murena e/os? If so, can you share compatibility issues you encountered? I'm eager to ditch Android and would like to anticipate which troubles may lie ahead. :)
Liebe Schwarmintelligenz, hat hier jemand das #Fairphone6 mit #murena #e/os und Erfahrungsberichte bzw Kompatibilitätsproblemen? Ich will wechseln und wissen welche Schwierigkeiten mich eventuell erwarten. :)
I used /e/OS before but don't currently, but here are my experiences, hope this helps:
1. Google Pay/Google Wallet is a lost cause, accept having to bring your bank card when leaving the house. 2. Anything you can find in IzzyOnDroid or F-Droid will work fine. I strongly recommend installing an F-Droid client: izzyondroid.org/quickstart/, you can already get used to some apps :) 3. Most apps will work fine but not all. Try Sapio and Plexus to see compat info. Mostly banking apps are a pain.
I think you do need to go in with a mindset and willingness to compromise sometimes. If you go in with the idea of "I can run anything I randomly download from Google Play" you will be disappointed. But if your general way of using your phone is "as little apps as possible" it may be fairly pain-free, especially if you prioritize apps from IzzyOnDroid and F-Droid.
May be worth making a list of apps you really aren't willing to lose and ask other /e/OS users specifically about those apps.
@SylvieLorxu Thanks a lot for your thoughts and the insights! I fully agree about the mindset. I also see it as an opportunity to declutter and/or widely review and replace what I've been using with FOSS apps. My only real need is the bare minimum for digital participation (public transport , banking seems to have mixed support for my banks 🫣, eID, etc), I'll celebrate everything beyond this as a nice add on
I sadly can't say much useful about public transport (I drive, sorry) although I do know a few people use my app Catima together with https://zügli.app/ for their Deutschlandticket (though it's broken in 2.41.0 - 2.41.3, version 2.41.4 with the fix isn't available everywhere yet).
Banking is very inconsistent yeah, depends on the bank. eID depends per country, though I think Germany is covered? f-droid.org/en/packages/com.go…
Hope some more Germans can fill in my missing knowledge :)
We must not stand in awe of the magnitude of our global contradictions. We need to identify the contradictions at hand and look for transiterative processes we can enact now to impact our trajectory, so that greater contradictions come to be at hand.
A friend of mine, Beqa Gozalishvili, a very talented developer from Georgia the country, announced an early stage of his #SAPI5 wrapper for the popular #ESpeakNG #TTS engine. bug reports and feature requests are welcome, he says in his Telegram channel. He does speak English. github.com/gozaltech/espeak-ng… #Accessibility #ScreenReader #Windows #JAWS #NVDA
@alexhall you can already download it. Pull the conversation if you can and check out the github page. I can get you a download link if you can't find it
@alexhall @Bri I installed it. Quite responsive. Problem though. And this isn't for my exact needs, but, there isn't an installable language selecter. Say someone speaks English and Spanish, they won't have Spanish installed unless there's a file I'm missing? I don't know
@alexhall @Bri Does it let you select voice/variant pairs to be exposed in SAPI like the old one did? Pretty sure he submitted this to ng issue number 7 but IDK if it's going to be pulled in.
@Bri @alexhall Oh. Meh. So I can't expose multiple variants, e.g. a Klatt one and a non-klatt one to use across given langs? Sad. Hopefully that can be added.
You may have heard that the maintainer of med-mastodon.com, a large, prominent Mastodon server, recently decided to shut down the server without warning, leaving his users with no access to their data. [EDIT: it's back up! Ref: mastodonapp.uk/@wild1145/11584…] This could happen to you at any time. If you care about this, you should be backing up your data regularly. Here's a post I just wrote about how to do that, and about what you can do with your backed-up data. #FediTips blog.kamens.us/2026/01/02/the-…
Your server or your Mastodon account could go away at any time. If you don’t want to lose your data, you should back it up. Here’s how to do that, and how to restore it later.
Collaborate for free with online versions of Microsoft Word, PowerPoint, Excel, and OneNote. Save documents, workbooks, and presentations online, in OneDrive. Share them with others and work together at the same time.
Stack Overflow's monthly question volume has collapsed about 300 -- levels not seen since the site launched in 2009, according to data from the Stack Overflow Data Explorer that tracks the platform's activity over its sixteen-year history.
Well, we've already taken to defending Wikipedia - it's gone from "it's just written by some random bod on the internet, it must be crap" to "it's written by actual human beings, it's great!"
Over the weekend I got Pi-hole up and running (first time user), and it sure is eye-opening seeing what your devices are up to.
The screenshot that's all red, for example, is my Samsung smart TV, which is constantly talking to Netflix (and other domains, even when using my PC and not interacting with the "smart hub".) I haven't opened the Netflix app in at least two weeks.
The other one all in green (allowed for now), is my Echo Dot with Alexa. That thing is talking to Amazon constantly...
Coincidentally I wrote a blog post about AdGuard Home (Pi-hole alternative) yesterday: blog.eischmann.cz/2026/01/04/a… (in Czech, but GT works well on it) I've also noticed how much the Netflix apps call home. Glad it's blocked now.
AdGuard Home: domácí ochrana nejen před reklamou Jak na blokování reklamy a rodičovskou kontrolu ve vlastní režii? AdGuard Home je na to výborné řešení. Používám ho už půl roku a v článku jsem se podělil o praktické zkušenosti z instalace a provozu, popsal jeho nejdůležitější vlastnosti a taky se trochu věnoval srovnání se známějším Pi-hole. #Adguard #AdguardHome #blokování #Pihole #reklama #soukromíblog.eischmann.cz/2026/01/04/a…(reakce na tento příspěvek se může zobrazit jako komentář pod článkem)
since Day 2 on CCCongress i am missing three of these definitively not an IED contraptions, designed to make music and deployed on the towel holders in three toilets, clearly marked with my DECT number and my nickname. If you took them and repurposed them for other shenanigans, i would love to have them back. As of now, i am out roughly 160 euros for replacements and that hurts.
Samsung screen reader users may have noticed that the TalkBack experience on One UI 7 and 8 can be frustrating when navigating recent apps. On the Recent Apps
Liebe #FediKirche, am Samstag, 21.2.2026 ist online das nächste #GottDigital #Barcamp . Wir wollten ja mal einen #Fediverse Workshop anbieten. Diesmal geht das relativ einfach, indem man sich am Anfang der Veranstaltung meldet. Ich habe bisher Zeit und würde mitmachen. Wer ist noch dabei als Teilgeber? Weitere Infos: gott.digital/ cc @alex @sannezc @onlinekirche
All theories of change are modes of transiteration, ways of "moving through again." Directing trajectories via action, feedback, and correction. Aristotle’s phronesis, Mao’s Mass Line, Freire’s pedagogy, Agile sprints. Power is control of correction. Trajectories are political.
My niece comes to the U.S. most years from Europe to visit her grandfather.
This year she was stopped at the airport, taken into an interrogation room, accused of being a prostitute, had her phone taken and scanned, and released.
You know, in case you were wondering why tourism to the U.S. is down.
Don't believe those who loudly claim email can not avoid metadata! They are ignorant of our continuous works on minimizing metadata:
DONE:
- no phone number other identifying data needed - no cleartext "Subject" - no cleartext "To" - randomized "Date" - no IP addresses - group/avatar/attachment/etc metadata only contained in encrypted message parts
Upcoming:
- servers to never see cryptographic ID metadata - remove "threading" and auxilliary headers - experiment with Sealed Sender
If in 2026 you make generic claims against the feasibility of using #email and #openpgp for secure decentralized messaging, there hardly is a way around taking a deeper look at RFC9850 and RFC9788 and the minimalist Rust-based approach chatmail.at takes. Hammering away at gpg and other crimes of how email-encryption doesnt work well ... frankly does not cut it for making a generic "it can not work" claim. See also chaos.social/@delta/1157966260…
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!
People generally use OpenPGP because they want to encrypt emails, so the topics intersect a lot, but it was about the challenges one faces when trying to use encryption with their mail client software.
If that's not what your team is doing, then the post doesn't apply to you, so there's no need to subtoot about it.
We'd be surprised if you find a "0day" exploit against standard #deltachat users but certainly prefer you finding something than eg putin/khomeini-aligned hackers.
we still welcome scrutiny even if you take the unusual path of directly publishing security findings without giving FOSS developers a chance to fix it for users. It's still better to become aware of issues now than in the future when there are bound to be many more users.
For decades, “The Year of the Linux Desktop” was treated as a tech world punchline. But as we enter 2026, the data shows the laughter has stopped. Between th...
Why in the world have you hashtagged this with #Accessibility? first, your video has not a single word about assistive technologies or accessibility, and second, people preaching Windows exodus usually keep silence on the small fact that most disabled people like blind, low-vision or low-mobility people have no viable, decent, professional and, the most important, accessible software replacement in the Linux world.
I appreciate your perspective. As a disabled creator myself, I use #Accessibility because I build my content to be accessible (Alt-Text, transcripts, etc.) and because my journey with Linux is tied to my own accessibility needs. I agree that Linux has a long way to go for many users with specific assistive tech needs. For me, the "exodus" is about reclaiming hardware that corporate EOL policies try to make inaccessible to those on fixed incomes.
Are you asking for proof of my disability? That is a deeply personal and invasive question. I do not share the specifics of my medical history or my personal assistive setup to qualify my lived experience or my work. Accessibility is a broad field, and my use of the tag stands on the standards I maintain for my content and the economic accessibility I advocate for.
You are gatekeeping a hashtag (which is silly) because you believe I am not disabled enough. That is wrong. I will not be replying to you again.
Oh no no, nothing personal, sorry for misunderstanding. If you are not comfortable about it, don't tell anything. I'm totally blind myself, hence asking. I see how blind people struggle in Linux environments, I struggle myself even in Windows with open-source solutions like Nextcloud and alternative office suites.
Getting very grumpy with people ghosting me over several #Accessibility fails I gave feedback for over the last month or so.
Emailed Winchester Christmas Market organisers because despite the event being listed as fully accessible, there was no way for a wheelchair user to actually approach 95% of the stalls there because they were up a step. Got an auto-mail stating the inbox was regularly monitored and they aim to respond to everything within 1-2 business days. No response. 3 weeks and waiting.
Emailed Showcase cinema because of poor quality viewing despite paying extra for fancy seats, the wheelchair accessible ones were right at the front and the image was warped and weird and the audio was atrocious. Plus one of the seats was broken and wouldn't recline. One response from a manager asking if the staff member we'd spoken to about the broken seat offered us anything (they didn't), no response since. 2 weeks and waiting.
Hey Guys, gals, people and pets, Satya Nadella is getting tired of people calling AI Slop output AI Slop. If everyone can stop that's be great as it is upsetting Microslops share holders.
Don't start calling Microslop Microslop or use the #microslop tag as it might burst the AI bubble and then Microslop might have to take SlopPilot out of Windows. As you know, that would be terrible.
yeeeeah, it might be time to to stop using GitHub for my projects. I know I said it last year but holy hell 🇺🇸 is not a safe place to be putting any of your data..
feels very relatable. I have half of my migration script done so I got the data at least local but since I want to either build my own custom forge software or cgit+mailinglist (less preferred option) its stuck. Wither on building the software or on me setting up all the lists in mailman...
(I don't want forgejo mainly because it 2 times self destructed on me to a point of impossible to recover. So I gave up on it for now. Codeberg I dont want since I dont want to land in another forge out of my control. I may set up codeberg mirrors for my more recent projects though)
1/2 Because I'm hearing a lot to the opposite effect and the discussion of it happens of course entirely in German and thereby without input from non German speakers: If you want the international community to engage with you and your topic, it is highly appreciated by many if you hold your talk in English. Be the live translation as good as it may, it does not invite to interact with you and your topic if your talk is in German. #39c3
1. No they won’t 2. They leave even when you don’t tax them 3. Things get better for most people when they leave 4. They say this so you won’t tax them
📣 Do-It-Blind (DIB) online Besprechung am Montag, 5. Jänner, um 19:00 Uhr. Du bist eingeladen! bbb.metalab.at/rooms/joh-szv-o… Wöchentlich am Montag besprechen wir neue Formen der digitalen und inklusiven Zusammenarbeit. Mach mit! 🛠️ #make #blind #inklusion
Learn using BigBlueButton, the trusted open-source web conferencing solution that enables seamless virtual collaboration and online learning experiences.
This is the first science post of 2026, and hopefully I will be able to do more this year. The Holidays were great and very awesome.
Always remember, you matter and to always be kind.
Today we are gonna be talking about The Integrated Circuit.
An integrated circuit (IC) is a miniature electronic circuit in which thousands to billions of electronic components such as transistors, resistors, capacitors, and diodes-are fabricated onto a single small piece of semiconductor material, usually silicon. Instead of building circuits with separate components, an IC combines them into a single, compact, and efficient unit. Because everything is built onto a single chip, ICs are smaller, faster, more reliable, and more energy-efficient than circuits made from discrete components.
, banking seems to have mixed support for my banks 🫣, eID, etc), I'll celebrate everything beyond this as a nice add on
雌雄揃って行動してることが多いように思うけど、まだ番になってない子なのかなぁ？
#misskey写真部 #photography #birdphotography #birds #野鳥
Landon
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦
in reply to Landon • • •PepperTheVixen ΘΔ
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦
in reply to PepperTheVixen ΘΔ • • •Devin Prater :blind:
in reply to PepperTheVixen ΘΔ • • •🇨🇦Samuel Proulx🇨🇦
in reply to Devin Prater :blind: • • •Devin Prater :blind:
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦
in reply to Devin Prater :blind: • • •Amir
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦 likes this.
Sean Randall
in reply to Amir • • •Do you know if codefactory are doing the same with their new android build?
Andre Louis
in reply to Sean Randall • • •James Scholes
in reply to Andre Louis • • •There is a 32-bit compatibility layer in the works for NVDA itself (although it currently only references SAPI4). But with any luck the need for every add-on to implement its own will go away.
github.com/nvaccess/nvda/pull/…
@cachondo @amir @fastfinge
Support for SAPI4 via a 32 bit shim runtime by michaelDCurran · Pull Request #19412 · nvaccess/nvda
GitHubSean Randall
in reply to James Scholes • • •🇨🇦Samuel Proulx🇨🇦
in reply to Sean Randall • • •Sean Randall
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦
in reply to James Scholes • • •James Scholes
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦
in reply to James Scholes • • •James Scholes
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •I see the "Secure add-on runtime" on the roadmap, with the note that "The first version of the runtime will provide support for speech synthesis and braille devices."
I don't see any implication that any 32-bit compatibility layer will only work for secure add-ons, which is hopefully a bit of a leap.
Still, the fact that people don't know what will or won't be happening, or whether their preferred synthesiser(s) will work or not, continues to be a big part of the problem. @cachondo @FreakyFwoof @amir
🇨🇦Samuel Proulx🇨🇦
in reply to James Scholes • • •Cleverson
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦 likes this.
Cleverson
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦 likes this.
🇨🇦Samuel Proulx🇨🇦
in reply to Cleverson • • •🇨🇦Samuel Proulx🇨🇦
in reply to Andre Louis • • •@FreakyFwoof@cachondo@amir You should be able to get either Gemini or Codex to help you, depending on what AI you have access to. The workflow would be:
1. download gemini-cli or codex-cli, and get them installed and configured.
2. clone all of the sourcecode from github.com/fastfinge/eloquence_64/3. Delete the tts.txt and tts.pdf files, so you don't confuse it with incorrect documentation.
4. Find any API documentation for orphius that's available, and add it into the folder.
4. Run codex-cli or gemini-cli, and tell it something like: "Using the information about how to develop NVDA addons you can find in agents.md, and the information about the Orphius API I've provided in the file Orphius-documentation-filename.txt, I would like you to modify the code in this folder to work with Orpheus instead of eloquence."
It will go away for five or ten minutes, ask you for permission to read and write the files it's interested in, and then give you something that mostly works. Now, build the addon, run it, and tell it about the errors and problems you have and ask it to fix them. In the case of errors, include the error right from the NVDA log, and for bugs and problems, tell it exactly what it's doing wrong, and exactly what you want it to do instead. Keep doing this until you wind up with a working addon.
Think of AI as a particularly stupid programmer, and you're the manager in charge of the project. You should be able to get this done without paying anyone.
GitHub - fastfinge/eloquence_64: Eloquence synthesizer NVDA add-on compatible with 64-bit versions of NVDA
GitHubjohann reshared this.
Andre Louis
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦
in reply to Andre Louis • • •Hamish
in reply to Andre Louis • • •Luis Carlos
in reply to Andre Louis • • •🇨🇦Samuel Proulx🇨🇦
in reply to Luis Carlos • • •Luis Carlos
in reply to Andre Louis • • •🇨🇦Samuel Proulx🇨🇦
in reply to Luis Carlos • • •🇨🇦Samuel Proulx🇨🇦
in reply to Sean Randall • • •Andre Louis
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦 likes this.
Chris Smart
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦
in reply to Chris Smart • • •Zach Bennoui
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •Really interesting article. I'm particularly passionate about this subject, I've been fascinated with TTS for a number of years. I've trained many voices, both for Piper and some of the newer LLM based systems, and while I can't speak to the speed issue, training data is extremely important.
What you feed into these models has a big impact on the voice's performance overall. If you give it stuff scrape from the web, random audiobooks that weren't optimized for TTS, things like that, you're not going to get good results for the type of work screen reader users do every day. This applies to all of these systems, not even just neural networks. The latency / responsiveness issue is something we'll have to solve at some point, because I don't think using TTS systems last updated in 2003 is going to work out in the longterm, as much as I love Eloquence.
In my ideal world, we would have either a machine learning based or formant system that is easy to train / maintain. Big companies have lost interest in on device TTS, not even just for screen reader users. Many of the solutions being put out now are cloud based, and while developers are still creating on device models, as said in the article, they're not optimized for our needs and may never be. I think we have to take matters into our own hands and figure this out, but I believe with enough people we can make it happen.
🇨🇦Samuel Proulx🇨🇦 likes this.
🇨🇦Samuel Proulx🇨🇦
in reply to Zach Bennoui • • •🇨🇦Samuel Proulx🇨🇦
in reply to Zach Bennoui • • •Zach Bennoui
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦
in reply to Zach Bennoui • • •Cleverson
in reply to Zach Bennoui • • •🇨🇦Samuel Proulx🇨🇦 likes this.
Chi Kim
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦
in reply to Chi Kim • • •Jayson Smith
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •🇨🇦Samuel Proulx🇨🇦
in reply to Jayson Smith • • •Paul L
in reply to 🇨🇦Samuel Proulx🇨🇦 • • •isn't it possible to "pregenerate" the speech with all the necessary IDs so that you can navigate and interrupt at will?
Just as one generates SSML from rich text (including maths formulas) before generating speech.
It would even be better to catch intonations, breaths and others, unchanged instead of letting the TTS generating a "pleasant full phrase" (a wrong expectation).
I find your post intriguingly close to the emerging reaction against the Ai-generated #mundaneslop
.
🇨🇦Samuel Proulx🇨🇦
in reply to Paul L • • •@polx Maybe, but probably not. Doing that would result in a lot of wasted resources generating text I'm never going to listen to. Think about the average user interface: dozens of menus, and toolbars, and ads, and comments, and so on. Plus, the text changes constantly, on even simple websites. That's not even taking into account websites that just scroll constantly. It might be possible to create some kind of algorithm to predict the most likely text I'll want next, but now we've just added another AI on top of the first AI.
I think a better solution might be to make the text to speech system run on different hardware from the computer itself. This is, in fact, how text to speech was done in the past, before computers had multi-channel soundcards. This has a few advantages. First, even if the computer itself is busy, the speech never crashes or falls behind. Second, if the computer crashes, it could be possible to actually read out the last error encountered. Third, specialized devices could be perhaps more power and CPU efficient.
The reason text to speech systems became software, instead of hardware, is largely because of cost. It's much cheaper to just download and install a program than it is to purchase another device. Also, it means you don't have to carry around another dongle and plug it into the computer.