# machine-learning
- u
(Dawn) Will Draw Fictional Women
07/19/2022, 4:00 PMno no there are some wrong ways - w
Waltman13 (semi-active)
07/19/2022, 4:09 PMI just make everything one sentence and put exclamation points at the end of almost every line of text. Probably not the right way to transcribe stuff but that usually works the best for my models somehow (mostly unless you're bubbles who's just a fucking nightmare to get working right). - m
matty1406
07/19/2022, 4:11 PMUnless the file I'm transcribing is one sentence only, i would not go for only one sentences... - m
matty1406
07/19/2022, 4:11 PMall my old, bad models were caused because of only one sentence, because it was smaller wavs and so wasn't good - g
gabrielck11
07/19/2022, 4:21 PMWe should make a voice changer on uberduck and you can sound like any character, it’s like talknet but live! Here’s an example I made - g
gabrielck11
07/19/2022, 4:22 PMIf we can use talknet but change some of the code to make it a voice changer it’ll work! - g
gabrielck11
07/19/2022, 4:22 PMDon’t worry about the reference audio button just say there’s a Voice Changer Button - u
{K EY1} (Kei)
07/19/2022, 4:23 PMDo you want it to be real time? If not, you can already have something like that by just recording the audio on your ipad or whatever device - u
{K EY1} (Kei)
07/19/2022, 4:23 PMIf you want it to be real time, talknet can't do that, but something like realtime yukarin can - u
(Dawn) Will Draw Fictional Women
07/19/2022, 4:23 PMyou have no clue how this works do you - g
gabrielck11
07/19/2022, 4:24 PMYea I meant real time, I forgot the word - u
{K EY1} (Kei)
07/19/2022, 4:26 PMYeah that won't work with talknet We'd need to implement realtime yukarin or something and that would be basically impossible because it requires you to train your own reference model with your own speech. You can't just use a universal model tmk. Other voice conversion models might work; tbh realtime voice conversion isn't my strong suit so i'm not 100% sure of the options If you can find something and demonstrate that it works, you can talk with zwf about implementing it but you need to think about how it would work in the site and how it would be implemented - u
(Dawn) Will Draw Fictional Women
07/19/2022, 4:26 PMjust wait for nyvox to hit steam - u
(Dawn) Will Draw Fictional Women
07/19/2022, 4:27 PMtheres no need for real-time on uberduck - u
{K EY1} (Kei)
07/19/2022, 4:28 PMThe benefit of realtime on something like uberduck would be sharing models on a streamlined platform but I have a feeling it would be quite difficult to implement - g
gabrielck11
07/19/2022, 4:28 PMWell Voicemod made AI voices so if one of us can find out how they made it we could use that for the characters on uberduck (ofc the person who made the model has to like make a voice changer voice) - u
(Dawn) Will Draw Fictional Women
07/19/2022, 4:28 PM…what - u
(Dawn) Will Draw Fictional Women
07/19/2022, 4:28 PMno???? - u
{K EY1} (Kei)
07/19/2022, 4:29 PMI don't know anything about voicemod so i'm going to leave that up to you - u
(Dawn) Will Draw Fictional Women
07/19/2022, 4:29 PMdont two of the voicemod voices sound like shit anyway - u
(Dawn) Will Draw Fictional Women
07/19/2022, 4:29 PMagain just wait for nyvox - u
{K EY1} (Kei)
07/19/2022, 4:29 PMWhat was nyvox again? - u
(Dawn) Will Draw Fictional Women
07/19/2022, 4:30 PMreal-time ai voice changer made by pathos - u
{K EY1} (Kei)
07/19/2022, 4:30 PMOh ok - u
{K EY1} (Kei)
07/19/2022, 4:30 PMYeah if that's easier to use than current options just wait - u
(Dawn) Will Draw Fictional Women
07/19/2022, 4:30 PMactually pretty good blatant got a free steam key for it - u
{K EY1} (Kei)
07/19/2022, 4:30 PMNice - u
{K EY1} (Kei)
07/19/2022, 4:31 PMI'd test realtime yukarin to see if there's any possibility of making a universal asr model for it but i don't have a dgpu lol - u
{K EY1} (Kei)
07/19/2022, 4:31 PMIsn't there also like stargan voice conversion