# machine-learning
- s
SidPlays_144p
07/15/2021, 7:10 PMkerchoo - u
unused
07/15/2021, 7:28 PM👁️🚗👁️ - u
user
07/15/2021, 7:28 PMYup that sums it up pretty well - w
Waltman13 (semi-active)
07/15/2021, 8:15 PMOk, working with isolated voice clips is a NIGHTMARE - w
Waltman13 (semi-active)
07/15/2021, 8:16 PMI've been trying to make 2 models and neither of them came out good. - c
Couch
07/15/2021, 8:19 PMyo uare using vocalremover right - c
Couch
07/15/2021, 8:19 PMare you clipping them then feeding to vocremover or do you feed to vocremover and then clip - w
Waltman13 (semi-active)
07/15/2021, 8:19 PMyou're talking about the google colab one right? - c
Couch
07/15/2021, 8:19 PMyes - w
Waltman13 (semi-active)
07/15/2021, 8:21 PMyes, I do use that one. as for the 2nd question, usually I just get the isolated audio from a youtube video then do the clipping in audacity. sometimes vise versa in case I can't get the video to be isolated due to some age restriction stuff. - c
Couch
07/15/2021, 8:34 PMupdate: reworked
regex and added comma to punctuation regex there is also a bug where the split regex stops working when an arpabet sequence is right after a word with no space i.e._arpabet_split_re
instead ofword{W ER1 D}word {W ER1 D} - w
Waltman13 (semi-active)
07/15/2021, 8:45 PMMaybe I shouldn't be making models using isolated samples. All they do is stroke out anyway - z
zwf
07/15/2021, 8:59 PMcurious if anyone here considers themself a particularly skilled GAN-based visual artist? - c
Couch
07/15/2021, 9:00 PMi havent tried any gans that have anythign to do with images - c
Couch
07/15/2021, 9:00 PMbut i have always been thinking about it ever since i attempted to download an entire comic strip series - i
IBob012
07/15/2021, 9:13 PMdamn i got a better result with jaiden and her clips are 100% vocal remover 5 isolated - c
Couch
07/15/2021, 9:15 PMive always used vr5 for most of my Data Sets - c
Couch
07/15/2021, 9:15 PMmarvin out of all of the mwas the most pain in the nuts to train because the transcript was wrong - c
Couch
07/15/2021, 9:19 PMthis one fixes aforementioned bug and also reworks the punctuation system onmaybe_get_arpabet - w
Waltman13 (semi-active)
07/15/2021, 9:35 PMI was able to get the model working at 80 epochs, but giving it too many words is still a no go for submitting as is still strokes. - w
Waltman13 (semi-active)
07/15/2021, 9:36 PM'ts gonna be a while until I get it working the way I want it to. - e
Emoto
07/16/2021, 6:12 AMThis be soundin pretty good, what model is this? - s
soap
07/16/2021, 7:38 AMIf someone finds a good episode I could use to make a voice, dm me - y
yuananf
07/16/2021, 7:49 AMIs there any tutorial of training another language model, for example French? Like a notebook or something? - y
yuananf
07/16/2021, 7:50 AMI see there are many other language models in https://github.com/mozilla/TTS/wiki/Released-Models, but the tacotron2 seem like implemented by Mozilla themself. - e
Emoto
07/16/2021, 8:36 AMTacotron2 could train any language aslong you offer a list based on the dialogue used in the wavs. depending on the language, it may be also a good idea to add a dictionary to help it understand what its working with quicker, since with english it's almost cookie cutter combining letters - e
Emoto
07/16/2021, 8:36 AMI've seen some very good models based on japanese already, so its defenitely possible - e
Emoto
07/16/2021, 8:37 AM - y
yuananf
07/16/2021, 9:24 AMThank you very much! - e
Emoto
07/16/2021, 9:30 AMno problem