Do you use a screen reader and read arabic content with it? Have you ever wondered why Arabic tts literally always sucks, being either super unresponsive, or gets most things wrong all the time? I've been wanting to rant about this for ages!
Imagine if English dropped most vowels: "Th ct st n th mt" for "The cat sat on the mat" and expected you to just KNOW which vowels go where. That's basically what Arabic does all day every day! Arabic uses an abjad, not an alphabet. Basically, we mostly write consonants, and the vowels are just... assumed? Like, they are very important in speech but we don't really write them down except in very rare and special cases (children's books, religious texts, etc). No one writes them at all otherwise and that is very acceptable because the language is designed that way.
A proper Arabic tts needs to analyze the entire sentence, maybe even the whole paragraph because the exact same word could have different unwritten vowels depending on its location, which actually changes its form and meaning! But for screen readers, you want your tts to be fast and responsive. And you do that by skipping all of that semantic processing. Instead it's literally just half-assed guess work which is almost wrong all the time, so we end up hearing everything the wrong way and just cope with it.
It gets worse. What if we give the tts a single word to read (which is pretty common when you're more closely analyzing something). Let's apply that logic to English. Imagine you are the tts engine. You get presented with just 'st', with no surrounding context and have to figure out the vowels here. Is it Sit? Soot? Set? Maybe even stay? You literally don't know, but each of those might be valid even with how wildly the meaning could be different.
It's EXACTLY like that in Arabic, but much worse because it happens all the time. You highlight a word like 'كتب' (ktb) on its own. What does the TTS say? Does it guess 'kataba' (he wrote)? 'Kutiba' (it was written)? 'Kutub' (books (a freaking NOUN!))? Or maybe even 'kutubi' (my books)? The TTS literally just takes a stab in the dark, and usually defaults to the most basic verb form, 'kataba', even if the context screams 'books'!
So yeah. We're stuck with tools that make us work twice as hard just to understand our own language. You will get used to it over time, but It adds this whole extra layer of cognitive load that speakers of, say, English just don't have to deal with when using their screen readers.

Coqui TTS: Free Text-to-Speech

https://coquitts.com

#HackerNews #Coqui #TTS #Free #Text-to-Speech #AI #Speech #Synthesis #OpenSource

Favorite thing lately is finding an article I wish were in podcast form, saving the text to a .txt file, then having TTS Util use RH Voice to convert the file into an audio reading, and listen to my own little robotic FOSS nanny read me the stories I want to hear in my headphones as I do yardwork.

An interactive demo for developers to try the new text-to-speech model in the #openai #tts #audio #api https://www.openai.fm

New feature out today in the #flash news ⁨#aggregator⁩: TTS (text-to-speech).

Two weeks ago (see: https://hachyderm.io/@paolog/114131812522555845) the platform acquired the capability to create multiple lists to save / bookmark articles. Building upon that function, it is now possible to have it read aloud all the articles in a list, using the local #TTS (text-to-speech) voices on your device.

This could be useful for #accessibility or to create a kind of "synthetic #podcast".

For example this short (1' 13") video shows a Samsung tablet reading three articles using the Samsung TTS voices (English and Italian). A light grey background shows the progress while each article is read. The tiny green toolbar at the bottom makes it possible to pause, skip forward and backward and stop altogether the reading.

https://youtube.com/shorts/5hx9WQzGgps

https://yewtu.be/watch?v=5hx9WQzGgps

Note: for this to work you need to install local voices (it does not work well with "cloud" voices), and disable the automatic screen locking (otherwise the reading stops abruptly when the screen locks).

I'm bringing back my TTS notification reader. Raspberry Pi 3 and Piper local voice.

Danke an Thorsten, dass er seine Stimme an uns alle verschenkt hat. #TTS #TextToSpeech

Sogar die low-Verwion der Piper Stimme klingt echt gut und läuft mit SherpaTTS auf meinem Handy.
https://github.com/woheller69/ttsEngine

https://netzpolitik.org/2025/text-to-speech-dieser-mann-hat-seine-stimme-verschenkt/

Here's another fun project! AIChat lets you generate conversations between two LLMs on any topic using OpenAI API. You can mix and match models from Ollama, MLX, Claude, OpenAI, Google AI Studio, etc. It uses Kokoro-ONNX for TTS. Check out an example + the GitHub repo:
https://www.youtube.com/watch?v=FgSZLZnYlAE
https://github.com/chigkim/AIChat
#LLM #TTS #AI #ML

Now there's MLX-Audio! They have Kokoro and CSM-1B for now. It's the same dev team for MLX-VLM, so I have high hope! #TTS #AI #ML https://github.com/Blaizzy/mlx-audio

It's talking like a human!

Have a call with Sesame's demo—their TTS model with fine-tuned Gemma3:27B

https://www.sesame.com/research/crossing_the_uncanny_valley_of_voice#demo

For everybody who uses the open source TTS (text-to-speech) application Piper: I have created a Docker file to run Piper directly within the container. No need to install the correct Python version. Everything comes pre-installed in the Docker image. Open source too: https://github.com/tderflinger/piper-cli-docker
#python #piper #tts #voice #docker #opensource

Going LIVE on YouTube at 7 PM Central! I need a system that reads my chat messages aloud in real time—so we're building one from scratch using AI tools! This is the perfect chance to get introduced to AI coding and see it in action.

Don’t miss out—tune in and be part of the build!

https://www.youtube.com/channel/UC7J5I0muqqQF51axXfPwrgA

#Python Urwid TUI tool for listening text spoken in different accents based on available espeak languages on the system. Generated with Claude 3.5 for ~$0.20. Idea felt too stupid to spend time writing manually but LLM worked quite well in this case. https://codeberg.org/jasalt/espeak-variator #terminal #audio #unix #tts

18F was doing exactly the type of work that DOGE claims to want – yet we were eliminated.

When former Tesla engineer Thomas Shedd took the position of #TTS director and met with TTS including #18F on February 3, 2025, he acknowledged that the group is the “gold standard” of civic technologists
https://18f.org/

By now, it's no longer a mystery what DOGE is up to - and it's not eliminating #FraudWasteAndAbuse.

Demo of new Siri voices in VoiceOver on iOS 18.4.