# tacotron-2-support
- t
Tails
12/30/2022, 1:51 PMjust concatenate the audios dont mess with the pronunciation? - t
tylerdurdenceketi
12/30/2022, 1:52 PMWdym with pronunciation? - t
tylerdurdenceketi
12/30/2022, 1:52 PMYou need to concat lines too - t
Tails
12/30/2022, 1:53 PMso i would end with about 200 filesm is that ok? - t
tylerdurdenceketi
12/30/2022, 1:54 PMIf you train on a base model it is ok. - t
Tails
12/30/2022, 1:55 PMoh, thats another question, i used the LJspeech 1 speaker female as base - t
Tails
12/30/2022, 1:56 PMi dont think so... - t
tylerdurdenceketi
12/30/2022, 1:56 PMUse portuguese base if you can - t
tylerdurdenceketi
12/30/2022, 1:57 PMIf there is no portuguese base, you need to create a base model first - t
Tails
12/30/2022, 1:58 PMwhen chose the portuguese base there is a msg about using the english base and works only with a lot of data - t
tylerdurdenceketi
12/30/2022, 2:01 PMYou can train a base model with Portuguese audiobooks etc. - t
Tails
12/30/2022, 2:02 PMi saw realy good portuguese models on uberduckAI, they used another base model than mine? - t
tylerdurdenceketi
12/30/2022, 2:02 PMWe trained a Turkish base with audiobooks, Youtubers etc. - t
tylerdurdenceketi
12/30/2022, 2:03 PMProbably You can try to transliterate Portuguese to English but it might have English accent - t
tylerdurdenceketi
12/30/2022, 2:03 PM@hecko do you have portuguese base - h
hecko
12/30/2022, 2:04 PMi do not - h
hecko
12/30/2022, 2:04 PMthere used to be a base but its creator removed it - t
tylerdurdenceketi
12/30/2022, 2:04 PM@Tails Well you need to create a base model - t
Tails
12/30/2022, 2:05 PMi see... the english model fits fast becouse of the learning transfer - t
tylerdurdenceketi
12/30/2022, 2:05 PMYou can train a Portuguese base on English base - t
Tails
12/30/2022, 2:05 PMso without the correct base model i cant fit my model with that small amout of data - t
tylerdurdenceketi
12/30/2022, 2:06 PMAbsolutely - t
Tails
12/30/2022, 2:07 PMwith many different speakers and different intonations? - t
tylerdurdenceketi
12/30/2022, 2:07 PMYes One female one male Or Many female and male voices - t
Tails
12/30/2022, 2:07 PMabout 1k of audios? - t
tylerdurdenceketi
12/30/2022, 2:08 PMMaybe 20000 - t
tylerdurdenceketi
12/30/2022, 2:08 PMTurkish base has 21 hours of data - t
Tails
12/30/2022, 2:08 PMoh, thats gonna take a long time 🥲 - t
Tails
12/30/2022, 2:09 PMhey i think i saw a portugues model somewhere - t
Tails
12/30/2022, 2:11 PMi just need the .pt model right?