# machine-learning
- u
(Dawn) Will Draw Fictional Women
07/30/2022, 1:03 PMi said it once ill say it again - u
(Dawn) Will Draw Fictional Women
07/30/2022, 1:03 PMw tech l motive - u
{K EY1} (Kei)
07/30/2022, 2:49 PMOk cool I'll test it just in case first - h
hecko
07/30/2022, 4:13 PMooh so they're converting a main voice into the custom one instead of synthesizing from scratch - h
hecko
07/30/2022, 4:13 PMwonder how useful that'd be for us - t
TheRoyalRuby2000
07/30/2022, 4:14 PMHey Hecko, my man I just wanted to say congrats on becoming site & server mod! - t
The Wiggles and High School D×D
07/30/2022, 6:54 PMI'm at Epoch 6 and wow, the loss is so much lower - t
The Wiggles and High School D×D
07/30/2022, 6:54 PMI might have a decent model with only a couple dozen epochs - a
Arienne
07/31/2022, 11:05 AMnever seen loss go this low so quickly before - a
Arienne
07/31/2022, 11:05 AM4 epochs - g
gabrielck11
07/31/2022, 11:05 AMArienne Grande - a
Arienne
07/31/2022, 11:08 AMinteresting dataset. only 100 lines, but 5 takes of every line. never seen a talknet or tacotron dataset with multiple takes of the same line before - a
Arienne
07/31/2022, 11:08 AMnot of every line i mean - a
Arienne
07/31/2022, 11:09 AMi already did a talknet, and it was picture perfect after only 50 spectro epochs with almost no audible artifacts for any input text i could think of. i wonder if talknet likes having more than one take - u
(Dawn) Will Draw Fictional Women
08/01/2022, 3:27 AMif they vary in timbre absolutely - t
tanooki426
08/02/2022, 2:35 PMI tried to convert an MP4 video of Scott Durbin to a WAV file to edit in Audacity and it sounded all glitchy! What should I do? - r
Radak
08/02/2022, 3:07 PMIs there a real voice cloning for singing AI‘s if so can someone please send me a link in private or here so that way - t
The Wiggles and High School D×D
08/02/2022, 5:18 PMAfter Epoch 26 I checked to see how my models function on both languages. The one with more audio is English text while the one with the blip at the end is romaji (actual Japanese is a bust). - t
The Wiggles and High School D×D
08/02/2022, 7:22 PMHow many epochs does it take before I start to get results that sound like the input speech? - t
The Wiggles and High School D×D
08/02/2022, 7:23 PMMy loss is almost always less than 1 - u
{K EY1} (Kei)
08/02/2022, 7:46 PMIf it sounds like this i'm a little worried you aren't training correctly (like maybe the cleaners are wrong, or the transcription is somehow off) - t
The Wiggles and High School D×D
08/02/2022, 9:46 PMWell before it seemed to be training incorrectly too - t
The Wiggles and High School D×D
08/02/2022, 9:47 PMI'm using basic_cleaners with ARPAbet - u
{K EY1} (Kei)
08/02/2022, 10:12 PMWhat do the japanese transcriptions look like? - t
The Wiggles and High School D×D
08/03/2022, 2:26 PMspeakers/0000_jvs001/parallel100/wav24kHz16bit/VOICEACTRESS100_001.wav|また、東寺のように、五大明王と呼ばれる、主要な明王の中央に配されることも多い。|0 - h
haru0l
08/03/2022, 2:48 PMproper Japanese won't work in tacotron without modification - h
haru0l
08/03/2022, 2:49 PMit recognises Japanese kanji characters as chinese ones by default - c
Couch
08/03/2022, 3:07 PMyeah but thats only because unidecode is very funny (which is used in english_cleaners for transliteration) - c
Couch
08/03/2022, 3:07 PMbasic_cleaners shouldnt have that problem - h
hecko
08/03/2022, 3:08 PMbasic_cleaners would delete everything because the charset doesn't have japanese in it