# machine-learning
- p
Pathos
06/30/2022, 12:25 AMIt absolutely does do cross lingual, though there it does impact quality a bit, lemme see if I had an example - u
{K EY1} (Kei)
06/30/2022, 12:26 AM👁️ - p
Pathos
06/30/2022, 12:27 AMA little bit of a quality impact. Someone asked how it sounded converting Hindi - u
{K EY1} (Kei)
06/30/2022, 12:27 AMOooo - m
mega b
06/30/2022, 1:05 AMsuper cool! \:D - o
Odcub Rodriguez
06/30/2022, 1:16 AMCAN ANYONE TELL ME WHAT THIS MEANS!?!? - g
Gosmokeless28
06/30/2022, 4:25 AMDoes it use uberduck's voice models? - z
zwf
06/30/2022, 4:38 AMit does not - z
zwf
06/30/2022, 4:38 AMbut the creators hang out in here, so I'm guessing they probably train on the same datasets - m
mega b
06/30/2022, 5:31 AMthe models don't really sound as good as they should - m
mega b
06/30/2022, 5:31 AMno offense to those lurking but dont hesitate to ask for suggestions \:-) - u
(Dawn) Will Draw Fictional Women
06/30/2022, 5:46 AMhesitate. hesitate a lot. - u
{K EY1} (Kei)
06/30/2022, 10:44 AMAre your checkpoints in
? This arose for me when i had my output path set for somewhere else and it couldn't find it Just move your latest checkpoint to/content/checkpoints/
and you should be good/content/checkpoints/ - a
aaq
06/30/2022, 2:27 PMany recommendations? like tacotron and talknet, or others i dont know about - u
{K EY1} (Kei)
06/30/2022, 2:28 PMRecommendations for what? - a
aaq
06/30/2022, 3:35 PMai - m
matty1406
06/30/2022, 3:36 PMtacotron for speech, talknet for singing and reference audio - u
{K EY1} (Kei)
06/30/2022, 3:36 PM^ - u
{K EY1} (Kei)
06/30/2022, 3:37 PMGenerally for speech you'll have a way better time with tacotron, though talknet can be used for normal speech too If you want to have a model that can sing / accept reference audio for inflection (or pitch) and timing, do talknet - u
{K EY1} (Kei)
06/30/2022, 3:38 PMIf you're preparing an english voice, though, you can do both since they both accept the same sort of dataset - m
MapleNeko
07/01/2022, 12:25 AMis there a calculation for how many batch size you should have? like 459 samples/dataset how many batch size should be what is the learning rate? a calculation might help - g
Gosmokeless28
07/01/2022, 3:28 AMI suspect that they have used my SpongeBob voice dataset to train a Tacotron 2 model of theirs. https://www.youtube.com/shorts/3r03UfFpVSY The voice strokes the same way that the June 25th version of my model does. - s
SpamZow (WoomyPearl)
07/01/2022, 11:15 AMA voice clone of a voice clone - g
gabrielck11
07/01/2022, 1:52 PMAlways wanted to do deepfakes but I think it’s a lot of work and you have to get nvidia cards - z
zwf
07/02/2022, 12:15 AM - c
Couch
07/02/2022, 12:16 AMincredible - z
zwf
07/02/2022, 12:16 AMshoutout to cryptocurrency for tanking - u
(Dawn) Will Draw Fictional Women
07/02/2022, 12:36 AMcommon crypto L - u
(Dawn) Will Draw Fictional Women
07/02/2022, 12:36 AM😎 - s
SpamZow (WoomyPearl)
07/02/2022, 1:09 AMFrom NFT seller to crypto detractor