Post ART6XCYXqgVMgjmENc by glowl@chaos.social
 (DIR) More posts by glowl@chaos.social
 (DIR) Post #ARSzw0sBixo80vqgl6 by hn50@social.lansky.name
       2023-01-09T12:35:06Z
       
       0 likes, 1 repeats
       
       Microsoft’s new text-to-speech model can duplicate anyone's voice in 3 secondsLink: https://mpost.io/vall-e-microsofts-new-zero-shot-text-to-speech-model-can-duplicate-everyones-voice-in-three-seconds/Discussion: https://news.ycombinator.com/item?id=34309306#microsoft
       
 (DIR) Post #ART6XCYXqgVMgjmENc by glowl@chaos.social
       2023-01-09T13:48:55Z
       
       0 likes, 0 repeats
       
       @hn50 A reason more to not make any noise before, someone calling you with an unknown number, hasn't introduced themselves. #ai #voice #scam #IdentityTheft
       
 (DIR) Post #ARTKrk7OAsIRvelQSe by joemo@mastodon.social
       2023-01-09T16:29:33Z
       
       0 likes, 0 repeats
       
       @hn50 open source tortoise-TTS has been able to do this for 6+ months now (maybe MSFT just forked it?), also a theoretical copy of DALL-E. The issue is not so much accuracy as how compute intensive (GPU intensive, really) it is to do the sort of careful mimicking, and with good prosody. Tortoise is ~5 seconds of a $1200 GPU to do one second of spoken text. https://github.com/neonbjb/tortoise-tts
       
 (DIR) Post #ARTRHydJQkCMCTaY3E by antygon@mastodon.online
       2023-01-09T17:41:35Z
       
       0 likes, 0 repeats
       
       @hn50 "Authorisation Picard - Alpha - Zero - One"