#dectalk

[dancio142]

I screwed around with some tts last night and came up with this masterpiece

[weezerfan68]

Weezer

[jansuta]

so someone ported my boy paul into utau

[cerulity]

Making your own Home Assistant (theorising)

Disclaimer: This is all theory and speculation. I have not tested anything or made my own home assistant yet, I just looked around for libraries and hardware that are likely compatible. I have not fully tested the compatibility or quality of these, this is simply the first iteration of an idea I have.

I just got news that Amazon Alexa has lost 10 billion dollars because their business model failed. This makes me happy, and has also made me realise that you can make your own home assistant.

Here are some of the links to things (I am aware that some are amazon, but it's the most global I could find. I encourage you to find other sellers, this is just what you should need. If you find anything cheaper or more local to where you are, go for it):

Hardware:

Raspberry Pi 4: https://www.canakit.com/raspberry-pi-4-2gb.html (RAM requirements may differ, I may do testing to see what comfortably runs)

8GB MicroSD: https://www.amazon.ca/Verbatim-Premium-microSDHC-Adapter-10-44081/dp/B00CBAUI40/ref=sr_1_3?crid=3CTN6X9TJXRR2&keywords=microsd%2Bcard%2B8gb&qid=1699209597&s=electronics&sprefix=microsd%2Bcard%2B8gb%2Celectronics%2C91&sr=1-3&th=1

Microphone: https://www.amazon.ca/SunFounder-Microphone-Raspberry-Recognition-Software/dp/B01KLRBHGM?th=1

Software:

Coqui STT: https://github.com/coqui-ai/STT

Coqui TTS: https://github.com/coqui-ai/TTS

If you have it set up correctly, you should be able to run both the STT and TTS in realtime (see https://github.com/coqui-ai/TTS/discussions/904).

After all of them are set up, the only thing to do is bridge it all together with software. There are bindings to Rust for both Coqui libraries (https://github.com/tazz4843/coqui-stt and https://github.com/rowan-sl/coqui-rs), and all that's left to do is implement parsing.

The libraries can also be swapped out for different ones if you like. If you can find and implement a DECtalk library that works for the Raspberry Pi, you can use that.

If I ever figure out how to manifest this idea, I will likely make the project modular so that you can use whatever library you want. You can even fork the project and include your own library of choice (if you can bind it to Rust).

Go FOSS!

[mothbagel]

like a normal person I have been obsessively researching the dectalk Wikipedia cat. The cats name seems to be tink. I love tink and their dct01 so much

[thisisntapainting]

[thothxv]

I'm going to hijack this thread slightly to talk about another one of Miku's forefathers.

Dennis Klatt was an early speech synthesis researcher at MIT. He did a lot of important things, but the one that's really recognizable is KlattTalk, better known as DECTalk. It sounds noticeably better and clearer than just about everything that came before it. It's still stiff and clipped, don't get me wrong, but it's really, genuinely better. If you want a sense of how much better, here's an audio recording of Dennis basically going through the evolution of speech synthesis, synthesizer by synthesizer. Dennis introduces each synthesizer, so you can get a sense of what his voice sounds like.

DECTalk's default voice, "Perfect Paul", was based on Dennis's voice directly. And while early units were expensive (one is pictured below with a cat for scale), later incarnations of DECTalk were more affordable, and the synthesizer snuck into all manner of places (you may know it as the voice from Moonbase Alpha, for example).

Dennis Klatt died in 1988 from thyroid cancer at the age of 50. In a cruel twist of fate, he began to lose his voice as the cancer developed.

But of course, in a way his voice lived on. One of the things Dennis had been interested in the accessibility applications of speech synthesis: creating ways for the blind to interact with the world, quite literally giving voice to the voiceless. And that's exactly what happened. The clipped, stiff, awkward, computer-generated rendition of Dennis Klatt's voice became someone else's.

This is the part I was omitting for dramatic reasons and that everyone who knows this story is probably screaming. A version of DECTalk's Perfect Paul served as Stephen Hawking's voice. And it was his voice. He went to such great lengths to keep it that when the original synthesizer was breaking down, an expert in hardware emulation was brought in to help develop a software version that would behave identically (the person in question was near, who you may know as the original developer of bsnes, although that is another, much more tragic story).

A dying man gave the very voice he was losing to a man who could no longer speak. There's something sad and bizarrely poetic about that.

whoever decided to turn daisy bell into a spooky dookie creepypasta song is fucking evil. that computer was brave enough to sing us a delightful little song and you do THIS to him? thats hatsune mikus grandpa dude. fuck you

[scummrevisited]

"two bodyguards drag me away" has been my fav way to finish insane sentences, its up there with "who said that", "can anyone hear me", "is this thing on" and "its so dark in here"

[miscpav]

DECtalk Flexidisc Demo 1984

[engvocaboost]

THE GLiTCHED ATTRACTiON by vocalo_shroomy ft. DecTalk Paul VCCV

THE GLiTCHED ATTRACTiON by @vocalo_shroomy ft. DecTalk Paul VCCV https://www.youtube.com/watch?v=q8_zDmnbyy0

This really does feel faithful to the TLT songs it's inspired by. It's wonderful.

[grrl-operator]

DECtalk DTC01 - 1984 Speech Synthesizer

[maribel-bat]

[sdinkbug]

gonna keep it for real I'm definitely objectum(?) but for those old janky ass TTS systems like DECtalk and SAM. especially DECtalk though because you can make it sing and its really fun and cute to fuck with the tone and make it all sassy

[alyssadotmp3]

this synthetic astronaut voice's singing genuinely means so much to me it always makes me so happy to see dectalk Paul remembered

[scummrevisited]

shoutout to friends btw. best thing earth has to offer