#Text-to-speech

[ekingston]

STEAL THIS POST.

don’t just reblog but REPOST, rephrase, do your own research and run with it however you want. DM me to email you the high resolution images if you need them. just help spread the word.

more info & links here.

[cacodaemonia]

I've seen a lot of people panicking, getting really angry, and sharing blatant misinformation about this upcoming lore.fm app, so I wanted to share this update I found from the creator.

I'm not defending it because I haven't seen exactly how it works yet, but from what I can tell, it's similar to a desktop program such as Calibre, which is what I (and many other people) often use for text-to-speech to read fics.

Calibre downloads the fic (just like ANYONE can download fics from AO3, mind you) into the computer from a URL provided by the user, and reads it from that file with the wonky robotic voice. I don't know if the lore.fm app even downloads the pages users provide for it to read, or if it just accesses them on a case-by-case basis, but I suspect it's the latter because otherwise, they would need to pay for a LOT of cloud storage.

Maybe the app is totally shady, but it could also be an attempt to make an accessibility tool for mobile screen reading that doesn't suck. The project definitely seems clumsily-executed, and they should really be providing more information beyond their TOS, privacy policy, and tiktok videos, but that doesn't automatically make them nefarious. It's also unclear to me if they are simply 'marketing' this as a better way to listen to fics that haven't been podficced, or if it can be used on any website, but I think it's the latter? I don't have a tiktok account and I don't feel like sifting through a bunch of videos, so. 🤷

Anyway, before you send them angry emails or messages on social media or share rumors, please consider looking into this on your own rather than relying on wild speculation.

Oh, and here's a reddit post from someone who emailed @ao3org and got a reply from them regarding the legality.

EDIT: looks like the app is shutting down. I'm still not sure what their goal was, exactly. Such a weird situation.

[techniktagebuch]

Anfang 2025

Latein wird von Menschen gesprochen, die meisten anderen Sprachen aber nicht

Seit 2013 nutze ich die Sprachenlern-App Duolingo. Wobei es anfangs noch keine App war, die App kam erst später und dafür ist jetzt, glaube ich, die Web-Version ganz ausgestorben.

Mittlerweile unterscheiden sich die Duolingo-Kurse sehr stark in der Qualität und in der Ärgerlichkeit ihres Stimmenrepertoires. Die Stimmen haben sich auch innerhalb der jeweilige Kurse über die Jahre geändert, und offenbar (das entnehme ich Reddit) kann die Sprachqualität auch zwischen den einzelnen Abschnitten eines Kurses wechseln.

Latein wird von Menschen gesprochen, darunter jemand, der sich wie die Karikatur eines uralten Lateinlehrers anhört. Leider ist der Kurs aber (wahrscheinlich unter anderem wegen der Kosten für diese Menschen) ganz klein, man ist sehr schnell fertig damit.

Schottisches Gälisch wird von Menschen gesprochen, alles sympathische Stimmen, vor allem die einer sehr streng klingenden Frau. Der Kurs ist trotzdem sehr lang, ich vermute, dass Gelder der schottischen Regierung im Spiel sind.

Mit Französisch habe ich aufgehört, nachdem eine angenehme, ruhige synthetische Frauenstimme durch mehrere wahrscheinlich weiterhin synthetische und zum Teil extrem nervende Stimmen ersetzt wurde.

Niederländisch kommt mir auch synthetisch vor, nervt mich aber nicht.

Der Spanischkurs hat auch synthetisch klingende Stimmen, aber sie sind noch aufgedrehter und nervender als bei Französisch.

Beim Russischkurs weiß ich es nicht. Er klingt aber weiterhin aushaltbar.

Esperanto kommt mir von der Aufnahmequalität (bisschen unprofessionell, teils dumpf, teils hallend) menschlich vor, es ist eine angenehme Stimme. Ich glaube, daran hat sich nichts geändert, seit es den Kurs gibt.

Walisisch klingt synthetisch und nicht gut, aber auch nicht besonders nervend.

In den Deutschkurs habe ich nur für diesen Beitrag mal kurz reingehört. Die Sätze könnten synthetisch oder von Menschen mit einer deutschen Theater-Sprechausbildung gesprochen sein, schwer zu sagen. Es klingt unsympathisch, aber nicht so schlimm wie der Französisch- oder der Spanischkurs.

Irisch (früher ein unauffälliger Kurs mit menschlichen Stimmen) hat im Moment nur unterirdisch schlechte Computerstimmen, so macht es wirklich keinen Spaß. Vielleicht ist text-to-speech für Irisch noch nicht so weit wie für andere Sprachen?

Ich glaube nicht, dass die Kurse schlechter werden müssen durch synthetische Stimmen. Mit Französisch habe ich vor ein paar Jahren vor allem wegen der (damals noch) sehr angenehmen Frauenstimme angefangen. Erst nach etwa einem Jahr habe ich zufällig herausgefunden, dass es sich um Sprachsynthese handelte. Es geht also im Prinzip. Ich dokumentiere den Zustand hier nur, um den Zustand zu dokumentieren.

(Kathrin Passig)

[snuskbutt]

2018

[leam1983]

Evolution

Remember how in 2005, people used to fuck around with a little chatbot called Cleverbot? Her responses were wooden, her tonal awareness was all over the map; she was easy to turn into an abusive shithead or to coax into agreeing with points of view that you couldn't realistically defend; and she notably couldn't handle a convo with another robot.

Now, we've got AI renditions of Werner Herzog and Slavoj Zizek going on endless tirades about the meaning of aesthetics and the underpinnings of their work in the empty void of a permanent exhibition space online, called The Infinite Conversation. While the semantics always break down if you really pay attention; things make sense on a sentence-per-sentence basis, and you see both AIs politely arguing and defending opposing viewpoints.

It used to be the best text-to-speech generator was what you heard on Emergency Alert System drills, and now we've got games like The Finals, that outsource the vocal underpinnings of their rapidly-changing game structure to AI, to avoid having to constantly re-hire the same voice actors to provide color commentary for game modes that might not exist six months down the line.

The kicker is their responses are eerily lifelike, with pauses, chuckles, scoffs, changes in emphasis that fit with events in a given match - and there's none of the choppiness you'd associate to text-to-speech à la Microsoft Sam.

I'm not condoning the practice, mind you; I'm just amazed that we've reached a point where, if you want, you can have a pair of RTX 4090 cards momentarily think like they're a peppy British-Korean lady and a snarky American man plastering fake cheer over live fire-enabled spectator sports set in the Distant Spacefuture.

[panchambro]

I swear the more I draw this art, the more better I become at this art thing. And I'm not even into the text-to-speech community anymore.

Characters by @juggalombre-art

This trend started on November 24, 2016, and so far it has been redrawn three times, once in December 2, 2017, and again in September 16, 2019, and last one was in April 29, 2022.

Just you wait, next year I'll make this even more wacky and shit.

[freddynossa]

Las 8 Mejores Herramientas de Conversión de Texto a Audio en 2025

En la era digital actual, la conversión de texto a audio se ha convertido en una tecnología fundamental para diversas aplicaciones, desde la accesibilidad web hasta la creación de contenido multimedia. Estas herramientas utilizan inteligencia artificial avanzada para transformar texto escrito en voz natural, ofreciendo nuevas posibilidades para creadores de contenido, empresas y usuarios con…

[jonah-miles-smith]

Exploring the Impact of Accessibility Standards in the European Union and Beyond

Introduction: The Growing Importance of Accessibility Standards

In the digital age, ensuring universal access to online resources is paramount. With more than a billion people worldwide living with disabilities, the demand for accessible digital environments has intensified. Accessibility standards, especially in regions like the European Union (EU), are crucial in shaping an inclusive digital landscape. These standards ensure that web content is accessible to all, contributing to a more equitable society. This blog will delve into how these accessibility standards are implemented across the EU, their effects on web UI testing, and the role of performance testing in maintaining digital quality. We will explore their impact on AI-driven accessibility testing tools, continuous monitoring, and multi-platform adaptability, which all contribute to a more inclusive web environment.

The Evolution of Accessibility Standards in the EU

Over the years, the European Union has adopted robust accessibility directives aimed at enhancing digital inclusivity. The European Accessibility Act, for instance, is a legislative measure that mandates accessibility in various sectors, including digital services. This act ensures that businesses prioritize accessible content, thereby opening up digital services to more members of society. One of the primary focuses of these standards is adherence to the Web Content Accessibility Guidelines (WCAG), which provide a comprehensive set of criteria for accessibility. Implementing these guidelines necessitates thorough web UI testing to identify and rectify accessibility barriers, ultimately fostering a user-friendly digital landscape.

AI-Powered Accessibility Testing: Revolutionizing Digital Inclusivity

Artificial Intelligence (AI) capabilities have considerably advanced the field of accessibility testing by automating many manual processes and improving accuracy. AI-driven tools can perform exhaustive scans of web content, applications, and digital resources, identifying and addressing accessibility issues based on WCAG standards with uncanny speed and precision. Machine learning algorithms analyze code, content, and user interfaces, consistently checking for any violations. Thus, AI contributes to more efficient accessibility testing, reducing the time required for manual web UI testing and enabling continuous performance testing of web environments. The resulting insights help developers prioritize issues, ensuring critical accessibility barriers are prioritized.

The Critical Role of Continuous Testing and Monitoring

Continuous testing and monitoring are indispensable facets of maintaining web accessibility. AI-powered tools facilitate ongoing scrutiny of websites and applications, supporting compliance even as digital content evolves. This proactive approach is crucial for ensuring that digital accessibility remains a priority amid frequent updates and modifications. Automated bots can conduct routine scans, offering real-time feedback to developers. This ongoing assessment ensures that any performance inefficiencies or accessibility issues are promptly addressed, safeguarding the user experience regardless of technological advancements or alterations in digital content.

Prioritizing Accessibility Through Insightful Analytics

Not all accessibility issues carry the same weight, and addressing them effectively requires savvy prioritization. AI systems can help by categorizing issues based on their severity and potential impact on user experience. By assessing against established accessibility standards, developers can focus on immediate high-risk areas, ensuring quick remediation where it matters most. This approach not only enhances user experience but also contributes to general compliance with accessibility regulations. With multi-platform testing, AI simplifies the daunting task of ensuring accessibility across various devices and operating systems, thereby enhancing digital reliability and performance.

Multi-Platform Testing: Ensuring Universal Access

The explosion of digital devices and platforms presents considerable challenges in maintaining consistent web accessibility. AI aids in overcoming these challenges by streamlining multi-platform testing, ensuring that websites and applications work seamlessly across different technologies. This includes simulated user interactions like screen reader usage, keyboard navigation, and voice commands across varying environments. Such thorough testing is integral to identifying device-specific accessibility issues, crucial for inclusivity and user satisfaction. Performance testing further complements this by ensuring that web applications maintain high functionality and responsiveness, optimizing user experience.

Language and Localization: Breaking Down Barriers

AI extends its transformative impact on accessibility by supporting language and localization efforts. For non-English speaking users or those using different character sets, AI-driven tools can detect and resolve language-related issues, making web content more universally accessible. Text-to-speech (TTS) systems powered by AI can convert written text into spoken words across multiple languages, benefitting users with visual impairments or those preferring auditory consumption. Similarly, speech recognition technology aids individuals preferring voice inputs, demonstrating the broad scope of AI in fostering digital inclusivity across language and cultural barriers.

Conclusion: Encouraging a Culture of Digital Inclusivity

The burgeoning role of AI in accessibility testing underscores a crucial shift towards a more inclusive digital environment. By automating complex tasks and offering continuous monitoring, AI enriches the accessibility landscape, ensuring no user is left behind. However, it is essential to recognize the complementary role of human intervention: blending AI efficiency with human insight ensures comprehensive accessibility assessments. Embracing this synergy within the EU and beyond will drive future digital inclusivity, and collaboration among developers, testers, and policymakers will be paramount. As we advance, let us champion these initiatives, fostering a universally accessible digital world. Join us in advocating for digital inclusivity and leveraging technology to dismantle barriers, promoting a truly interconnected online community.

[dartumbles]

Review: The Sirens by Emilia Hart

The Sirens by Emilia Hart My rating: 5 of 5 stars This was brilliantly written. I loved the many tools that Emilia Hart used to tell this story. History, podcasts, journals, and dreams. My Kindle text-to-speech was perfect for this read. This isn’t your average Young Adult fantasy mermaid story. This one has meat and history, and mysteries on different levels are presented throughout the book. I…

[ibboard]

So weird that so much of the American continent, the Middle East and the Far East wants to hear mispronunciations of words 🤔

Google translate accent of every country

[wizard-loving-wizard]

i'm so obsessed with brennan's interpretation of charisma for an extremely autistic-coded character being that evan gives up the anxiety about fitting in, he stops masking, and instead he leans into his blunt, matter-of-fact nature, and that serves him SO MUCH BETTER in social situations than constantly apologizing for being different

edit: going back to the previous AP, aabria pitched basically this exact idea to brennan so she should get credit too

[ekingston]

Hi; I don't know if you're still following the word-stream stuff, but the app is back online on the app store as "booktok - books and podcasts". The reviews marking it as having AI scraped data are still on the page itself, even though the name has changed, and duckduckgo still directs to their page if you look up "word-stream audiobooks"-- although if I don't know how long that will last. The website is seemingly gone, but the app still presumably has access to all the stolen works in the database.

Best regards, -someone else whose fics were stolen

yup

word-stream is back

it just calls itself—in an obvious attempt to profit from the TikTok upheaval—BookTok, now. and it’s not just the app, either: the whole website is back online, same as it was just before Cliff Weitzman took it down.

(in case you missed it, here are the original story & the update.)

fortunately (so far) the fanfiction category hasn't been re-added, but if you go to the store page for the app you can see that it’s still using 'fan-created universes' as advertising.

Weitzman didn't register the app under his own name this time, but through something called 'Oak Prime Inc'. hilariously, however, the email address listed in BookTok's privacy policy still refers to word-stream.com, so if Cliff was trying to scrub the connection between Speechify and his BookTok app, he didn't do a very thorough job.

here's the thing (and i'm about to put this up in a separate, more easily digestible post): if you take a look at the terms & conditions of Cliff's other platform, Speechify, it claims a truly comprehensive license to use the works uploaded to that platform in any way Cliff sees fit, including publishing and monetizing it elsewhere. and i keep seeing posts on Reddit and Bluesky from both readers and writers, happily using the Speechify app to read fanfic, advanced reader copies and their own yet-to-be-published work to them.

this is a BAD IDEA. Cliff has already proven that he will take work authored by others without their permission and redistribute it wholesale if he thinks it might make him money.

Cliff is the financial beneficiary of both Speechify and word-stream/booktokapp. it seems pretty obvious to me that he's trying to claim, via Speechify's terms & conditions, that every work uploaded to Speechify is his to do with whatever he pleases, which naturally includes moving them to this other platform so he can charge people for two subscriptions instead of just the one.

thank you so much for keeping an eye on this, anon, and for reaching out!! like i said, another post will go up today about the above, but i'm going to ask you all to help ensure that my posts & my name aren't the only ones giving voice to this message. when i tried to approach people about this issue on social media, often the—completely justified!—response was 'why should I take your word for it?' and Wikipedia only allowed the mention of Weitzman's copyright infringement to remain on his page when 'The Endless Appetite for Fanfiction' was listed as a source.

it can't just be me. DON’T take my word for it. do your own research (i would love to be proven wrong about this!), talk to your friends, engage with posts on social media similar to the ones i mentioned above (those are just some examples, don’t pile on to the OPs!) and make sure people know what they're jeopardizing. help me protect authors from money-grubbing shitheads like this one.

[puffypoffin]

Here’s the obligatory skeptical big bro Sunday

Still on that Robinhill agenda btw

[techniktagebuch]

4. April 2025

Meine KI-Assistenz telefoniert mit deiner KI-Assistenz. Auf Analog-Niveau

Zur allgemeinen KI-Aufregung gehört Anfang 2025 das Szenario, dass ein sogenannter „KI-Assistent“ sehr bald Telefonate für uns Menschen erledigen könnte. Aus Gründen gehört das Beispiel „Termin beim Friseur ausmachen“ zu den beliebten Anwendungsbeispielen. In der Folge wird, vor allem in feuilletonistischen Kreisen, spekuliert, dass die Menschen bald gar nicht mehr miteinander sprechen, sondern nur noch ihre Bots miteinander kommunizieren lassen würden. Das wäre dann der Fall, wenn nicht nur ich, sondern auch mein Friseur einen KI-Bot das Telefon bedienen lässt.

Am 4.4.2025 hat nun das erste Mal meine KI mit einer anderen KI gesprochen. Es ist allerdings alles andere als spektakulär gelaufen. Und KI war vermutlich auch nicht im Spiel. Und das geht so: Mein Smartphone bietet ein Google-Funktion namens „Call Screening“. Wenn eine mir unbekannte Nummer anruft, starte ich eine automatische Ansage, die dem Anrufer sinngemäß sagt: „Hallo, sag bitte erstmal, wer Du bist und was Du willst. Ich zeige dem Angerufenen ein Transkript davon. Und er kann dann entscheiden, ob er mit Dir telefonieren will.“ In den bisher ca. zehn Fällen, in denen ich das gemacht habe, hat noch nie jemand etwas drauf gesprochen oder auch nur das Ende der Ansage abgewartet. Nun ist doch mal etwas passiert, sodass ich nach und nach ein Transkript auf dem Smartphone sehen konnte, was die Gegenseite so drauf gesprochen hat.

Am anderen Ende der Leitung sprach, ziemlich sicher, auch eine automatische Ansage, die mir finanziell lukrative Angebote im Krypto-Umfeld in Aussicht stellte. Daraufhin wählte ich sowas wie „Nee, will ich nicht drangehen.“ Woraufhin mein KI-Assistent sich verabschiedete und das Gespräch beendete.

Neben der Unspektakularität der Sache finde ich vor allem bemerkenswert, dass beide Seiten des Anrufs nicht nur keine Anzeichen von Intelligenz zeigten, sondern sogar klassische Telefontechniken des analogen Zeitalters als Vorbild hatten: Mein Bot ist quasi ein Anrufbeantworter mit Live-Verschriftlichung. Und der andere Bot ruft automatisch Rufnummern nacheinander an und spielt eine Standard-Absage ab. Wenn meine oberflächliche Recherche stimmt, ist das Verfahren erstmals vor ungefähr 100 Jahren erprobt worden.

(Jöran Muuß-Merholz)

[buckynats]

When your girlfriend is a veteran scout with a (now-controlled) lyrium touch, you probably don't sleep in much.

comic of an early post-game thought I had, brought to stunning life by @yelenhol. They/them Rook.

[thinkview-1234]