# machine-learning
- t
TheTiagoFire
05/31/2022, 1:29 AMI see - t
TheTiagoFire
05/31/2022, 1:35 AMpromising - t
TheTiagoFire
05/31/2022, 1:36 AMah shit the pronunciation is off - u
{K EY1} (Kei)
05/31/2022, 1:36 AMhow off? - t
TheTiagoFire
05/31/2022, 1:37 AMthis off - u
(Dawn) Will Draw Fictional Women
05/31/2022, 1:38 AMpronunciation being incorrect is pretty par for the course - t
TheTiagoFire
05/31/2022, 1:39 AMI'm assuming there's a way to finetune it? - u
(Dawn) Will Draw Fictional Women
05/31/2022, 1:39 AMno? - t
TheTiagoFire
05/31/2022, 1:39 AMoh - u
(Dawn) Will Draw Fictional Women
05/31/2022, 1:39 AMif you mean hifigan finetuning yea thats a thing but that does nothing for pronunciation - u
(Dawn) Will Draw Fictional Women
05/31/2022, 1:40 AMthough if you really wanna be a stickler about it you can retrain with an arpabet transcript - t
TheTiagoFire
05/31/2022, 1:40 AMlike - t
TheTiagoFire
05/31/2022, 1:40 AMstart from 0 starting at the transcripting part? - t
TheTiagoFire
05/31/2022, 1:41 AMjust with different formatting? - u
{K EY1} (Kei)
05/31/2022, 1:41 AMnono you just train with a different notebook - u
{K EY1} (Kei)
05/31/2022, 1:42 AMgranted, the arpabet notebook is broken rn... so the only other option is the pipeline notebook - t
TheTiagoFire
05/31/2022, 1:42 AMyes but further training on what I've trained so far or is it re-training - u
{K EY1} (Kei)
05/31/2022, 1:43 AMretraining - t
TheTiagoFire
05/31/2022, 1:44 AMdang - t
TheTiagoFire
05/31/2022, 1:44 AMalrighty - t
TheTiagoFire
05/31/2022, 1:44 AMwouldn't it be better to always go with pipeline then if it's more reliable? - u
{K EY1} (Kei)
05/31/2022, 1:45 AMwell, pipeline is awkward it's a multispeaker notebook, and it's set up way differently from default tacotron i usually train my models on default, arpabet, and pipeline, then choose which one sounds the best - t
TheTiagoFire
05/31/2022, 1:46 AMoooh I see - u
(Dawn) Will Draw Fictional Women
05/31/2022, 1:55 AMi really dont trust autoarpa converters - u
(Dawn) Will Draw Fictional Women
05/31/2022, 1:56 AMjust get the transcript converted and manually fix it using the cmu guide - t
TheTiagoFire
05/31/2022, 2:01 AMwould it work if it's not standard english though - u
{K EY1} (Kei)
05/31/2022, 2:01 AMIt works with any dialect - t
TheTiagoFire
05/31/2022, 2:03 AMhow would I go about this? - t
TheTiagoFire
05/31/2022, 2:03 AMalso, just spotted this