# machine-learning
- u
{K EY1} (Kei)
06/03/2022, 7:08 PMHi gero 😰 - a
Awesome Face (GTX 1660 Ti)
06/03/2022, 7:09 PMoh boy... 👀 - g
Gero
06/03/2022, 7:09 PM[off-top] I've said a thousand times that I'm only 16 and I just love girls my age. I apologized many times, but you just blacklisted me xd - u
{K EY1} (Kei)
06/03/2022, 7:09 PMYou never said that, no clue what you're talking about now - u
{K EY1} (Kei)
06/03/2022, 7:10 PMFrankly idc anymore I just don't want you back in that one server, you caused too much trouble Sorry - u
{K EY1} (Kei)
06/03/2022, 7:11 PMAnyways if you want to use nnsvs with a russian dataset, the github with the hed and dict, and usage instructions, is up there I hope that helps - g
Gero
06/03/2022, 7:11 PMcan you unblock me and we can just talk about it again? I can even show a passport proving my age. - g
Gero
06/03/2022, 7:11 PMSorry for off-top - u
{K EY1} (Kei)
06/03/2022, 7:11 PMAlright it wasn't your age i was concerned about - u
{K EY1} (Kei)
06/03/2022, 7:11 PMLet's go to dms - g
Gero
06/03/2022, 7:25 PMSceryp, so to speak, is a little dumb. His standard is not perfect at all. And speaking of him as a person, he called my mother a dead whore and didn't apologize... - h
hecko
06/03/2022, 7:40 PMi once suggested openutau for talknet - u
{K EY1} (Kei)
06/03/2022, 8:12 PMOpenutau isn't compatible with talknet, but it can generate reference audio using other compatible voice models I often use vocal synthesizers to generate reference audio when i don't feel like recording on my own - h
hecko
06/03/2022, 8:12 PMno i mean like - h
hecko
06/03/2022, 8:12 PMhook it up directly - h
hecko
06/03/2022, 8:12 PMgenerate the phonemes and pitches instead of synthesizing a vocal to use as ref audio - h
hecko
06/03/2022, 8:12 PMi've done the latter with synth v and it feels like a waste - u
{K EY1} (Kei)
06/03/2022, 8:13 PMYeah, i've heard the hed doesn't have the right grouping I've never made a hed before though so idk And he said that about your mom?? Jeez... - u
{K EY1} (Kei)
06/03/2022, 8:14 PMIf you want to make a new russian hed, the people in the nnsvs server can probably help you There are hed creation tools, too - u
{K EY1} (Kei)
06/03/2022, 8:15 PMI'm not sure that would work I don't know a lot about talknet. Does it accept some sort of direct pitch control as opposed to using its asr model directly? - h
hecko
06/03/2022, 8:21 PMyeah controllabletalknet just converts the audio to pitches - h
hecko
06/03/2022, 8:22 PMyou'd have to have some way to generate consonant lengths though - u
{K EY1} (Kei)
06/03/2022, 8:23 PMSee that's the problem It uses asr for that part And unless there's some looparound way to make the piano roll input output info to the asr which outputs info to synthesize the talknet voice It wouldn't be feasible - u
{K EY1} (Kei)
06/03/2022, 8:24 PMIf you wanted to try it, i'd say go for it I'm sure stakira would be interested in anything you come up with But it sounds like a significant amount of work - h
hecko
06/03/2022, 8:37 PMthen just,, don't use asr for that part - h
hecko
06/03/2022, 8:37 PMrip it out - h
hecko
06/03/2022, 8:37 PMidk how to do it but it's gotta be possible - h
hecko
06/03/2022, 8:37 PMsomeone put it in there so we can put it out of there - c
Carson
06/03/2022, 10:45 PMi changed the model slightly and now it looks slightly less artsy but im still terrified - c
Carson
06/03/2022, 10:46 PMmaybe