Conversation

Notices

1. Embed this notice
   Karl Baron (kalleboo@bitbang.social)'s status on Saturday, 06-Apr-2024 04:42:31 JST Karl Baron

   They never got VOCALOID to do English well but whatever TTS Neuro-sama is using has cracked it.

   The job of a singer is over.

   • Embed this notice
     SuperDicq (superdicq@minidisc.tokyo)'s status on Saturday, 06-Apr-2024 04:42:31 JST SuperDicq

     in reply to

     @kalleboo@bitbang.social There are people who using Miku's voice to train AI models to make a "better" and more human sounding Miku that can actually speak English.
     
     I personally think this is heresy. The bad English is part of the Vocaloid aesthetic I say.

   • Embed this notice
     SuperDicq (superdicq@minidisc.tokyo)'s status on Saturday, 06-Apr-2024 04:46:19 JST SuperDicq

     in reply to

     @kalleboo@bitbang.social Also I'm pretty sure Neurosama just takes the vocal sound from the original song and just puts a (AI generated) filter on top of it to make it sound like Neurosama is singing it.
     
     She can't sing songs from scratch without having a human singer to "learn" from.
     
     Vocaloid is different because you can make songs from scratch using Vocaloid, as Vocaloid is more an "instrument" instead of a "filter" if you can even use those words in this context.

   • Embed this notice
     SuperDicq (superdicq@minidisc.tokyo)'s status on Saturday, 06-Apr-2024 04:50:25 JST SuperDicq

     in reply to

     @kalleboo@bitbang.social Sometimes little mistakes or character things that the original singer of a song does you can hear back in Neurosama's cover which kinda makes it unnatural, because nobody covers song with the same kind of precision, copying the original singer's flaws without adding any of your own.
     
     However Neurosama singing did cost the author a lot of work. You can hear Neurosama's voice algorithm being worked on with every karaoke video they make. The early ones sound really bad (almost funny) in comparison to this.

   • Embed this notice
     SuperDicq (superdicq@minidisc.tokyo)'s status on Saturday, 06-Apr-2024 04:51:20 JST SuperDicq

     in reply to

     @kalleboo@bitbang.social I don't think believe most vtubers are just random dudes with voice changers tho, most of them have a real voice.

   • Embed this notice
     Karl Baron (kalleboo@bitbang.social)'s status on Saturday, 06-Apr-2024 04:51:21 JST Karl Baron

     in reply to

     • SuperDicq

     @SuperDicq Yeah I think the neurosama singing is basically an advanced version of the “make a Donald Trump voice say this thing” voice changers, but that still means that you can replace all the commercially viable conventionally attractive pop singers with cute anime grills backed by some random dude singing in his basement (which is what all vtubers are already anyway am I right)

   • Embed this notice
     SuperDicq (superdicq@minidisc.tokyo)'s status on Saturday, 06-Apr-2024 04:53:40 JST SuperDicq

     in reply to

     @kalleboo@bitbang.social I do miss original Neuro-sama singing tho.
     
     It was really buggy and inaccurate to the point of actually being hilarious.
     
     https://yewtu.be/watch?v=Msm_Vasv0kA

   • Embed this notice
     Karl Baron (kalleboo@bitbang.social)'s status on Saturday, 06-Apr-2024 04:53:41 JST Karl Baron

     in reply to

     • SuperDicq

     @SuperDicq Yeah I was joking lol

   • Embed this notice
     Karl Baron (kalleboo@bitbang.social)'s status on Saturday, 06-Apr-2024 05:03:51 JST Karl Baron

     in reply to

     • SuperDicq

     @SuperDicq Haha that’s great

   • Embed this notice
     SuperDicq (superdicq@minidisc.tokyo)'s status on Saturday, 06-Apr-2024 05:03:51 JST SuperDicq

     in reply to

     @kalleboo@bitbang.social So yeah you can see all the progress that they made getting Neuro to sound natural.