# machine-learning
- b
bericks
05/28/2021, 3:37 PMyea - b
bericks
05/28/2021, 3:37 PMboth instrumental and vocals - d
Dyno
05/28/2021, 3:37 PMUnfortunately, the uberduck.ai bot has not gotten verified by Discord yet even though it should have been by now. It has been taking a long time for Discord to verify the bot for some reason. - b
bericks
05/28/2021, 3:37 PMtf you want - b
Balloonya from Pokémon
05/28/2021, 3:41 PMAny program can reduce the background noise from audio files? - j
Justin
05/28/2021, 3:52 PMCheck #841437191073955920 - a
Awesome Face (GTX 1660 Ti)
05/28/2021, 3:59 PMalright gave dino voice at least 3 hours on training with over loss 0.033 - u
user
05/28/2021, 4:16 PMHow can I fix pronunciation of my turret? - b
Balloonya from Pokémon
05/28/2021, 4:19 PMOnly vocals. - h
HoodyP
05/28/2021, 4:35 PMUm hey guys so... Can someone give me an example of how do I apply Arpabet into a dataset's transcript text file? Cuz I feel like the way I inputted my Arpabet data into the transcript is kinda wrong... - h
HoodyP
05/28/2021, 4:37 PMLike if I inputted normally as "wavs/1.wav| Hello World!", how should I replace it with arpabet? Bc my method of doing like "{H EH1 L OW1} {W ER1 L D}" seems like it doesn't work that nicely... - p
Pikachu ✓
05/28/2021, 4:38 PMWhy would you want to use Arpabet? - h
HoodyP
05/28/2021, 4:38 PMSo that I could improve pronunciations of my voice models... - h
HoodyP
05/28/2021, 4:39 PMLike applying this to my voice models' update and future models - h
HoodyP
05/28/2021, 4:40 PMI just need an example of how they write down the transcript text file with Arpabet - h
HoodyP
05/28/2021, 4:44 PMI'm gonna use that method I guess, combining all of the words into one bracket, separated with "." and removing numbers... - h
HoodyP
05/28/2021, 4:44 PMI'm just gonna do a lil more experimenting bc I don't even know how things work lol - a
Awesome Face (GTX 1660 Ti)
05/28/2021, 4:44 PMok sure - p
Pikachu ✓
05/28/2021, 4:45 PMthere is a reason for the numbers - h
HoodyP
05/28/2021, 4:45 PMThen maybe I'll keep the numbers - h
HoodyP
05/28/2021, 4:45 PMIdk lol - a
Alexius08
05/28/2021, 4:46 PMThe numbers are needed to denote the stress (or emphasis) on each syllable. 1 is for primary stress, 2 is for secondary stress, 0 is for no stress. - h
HoodyP
05/28/2021, 4:47 PMI know that tbh... but bc there is not much of a proper guide of how ppl prepare the transcript text file with Arpabet inputs to be trained with, I got a bit confused... - h
HoodyP
05/28/2021, 4:48 PMIf there is only such a guide... - u
user
05/28/2021, 4:56 PMHow much audio clips I should use to make it sound good? - h
HoodyP
05/28/2021, 4:57 PMAt least over 30 files or over 5 minutes... The more, the better - u
user
05/28/2021, 4:58 PMI made one with 40 files and 20 minutes of training, and it's very bad :c - u
user
05/28/2021, 4:59 PMI'll try put more files - p
Pikachu ✓
05/28/2021, 4:59 PMI have used 60 multiple times - p
Pikachu ✓
05/28/2021, 4:59 PMAlso, use RAW_INPUT in your colab synthesise notebook