Introducing TTS WebGPU: The first ever text-to-speech web app built with WebGPU acceleration! 🔥
High-quality and natural speech generation that runs 100% locally in your browser, powered by OuteTTS and Transformers.js. 🤗 Try it out yourself!
Demo + source code below 👇
Happy Thanksgiving!
the token needs to appear many times in the dataset during tokenizer training, but never during training. So start spamming your token somewhere singular and hope that it’s not used during training. Solidgoldmagikarp was on reddit count counting to infinity twitter.com/SoC_trilogy/...
The whole platform is kind of designed around it. See ATProto docs.bsky.app/docs/advance.... When publishing to a public social network, seems like individuals should have an understanding that it’s inherently public
Thing 5: probably most importantly: double check, asking details of shared secret knowledge before taking any extreme action. (What was your childhood stuffed animal?)
Thing 4: for those who it’s professionally practical for, use separate online personas without revealing actual personal details. YouTube personas could perhaps adopt fake personas instead of using real identifying characteristics.
Thing 3: look for inconsistencies in emotional tone from the actual person
Thing 2: at least for now, absence of ambient noises (breathing, non verbal sounds, background noises)
Thing 1: At least for now, the choice of language and the voice model are separate, so this might be a useful clue to pay attention to word choice
Had a dream that I was being scammed over the phone by a deep fake of my cousin. She was impatient and anxious. The real cousin was in the room with me. Weird. Got me thinking about the different ways of detecting fakes
Wow, those are some fabulous outfits
This was supposed to go in this thread bsky.app/profile/rnge...
That’s how they getcha
🙇
Wat why
