# tacotron-2-support
- c
Cris140
08/22/2022, 2:56 AMWelp, sorry to inform that but I think you lost the model - c
Cris140
08/22/2022, 2:56 AMTo train another language on legacy tt2, you need to change symbols and use basic_cleaners - c
Cris140
08/22/2022, 2:57 AMIn both training and synthesizing - c
Cris140
08/22/2022, 2:57 AMIf it's Pipeline the model probably can still be usable - u
{K EY1} (Kei)
08/22/2022, 2:58 AMi've done this before plenty of times, and i'm a developer in echelon's server because of my implementation of japanese. so i do know what i'm supposed to do to train a model in another language. thank you for your help, though - c
Cris140
08/22/2022, 3:00 AMWhich fork are you using? - u
{K EY1} (Kei)
08/22/2022, 3:04 AMto train, i'm using my custom notebook because i wanted to warm start on the french model (since it also uses ipa) - c
Cris140
08/22/2022, 3:04 AMOh, I'm using IPA as well, but the fork is private? I could take a look - u
{K EY1} (Kei)
08/22/2022, 3:05 AMit's this i've never had issues with it before, maybe this model just corrupted? https://github.com/Mildemelwe/Non-English-Tacotron-2-Training-Notebook - c
Cris140
08/22/2022, 3:08 AMOh I see the problem - c
Cris140
08/22/2022, 3:08 AMWhich notebook you're using to synthesize? - c
Cris140
08/22/2022, 3:11 AMThe problem is that it's expecting IPA symbols and the synthesizing part is probably sending just plain text, that's why it outputs nothing - c
Cris140
08/22/2022, 3:14 AMThat's just my theory, I would need to check the synthesizing notebook to be sure - u
{K EY1} (Kei)
08/22/2022, 5:47 PMI'm using this to synthesize I'm just really confused because i had zero problems with my french model https://colab.research.google.com/drive/1NVA3ndxhYWsKn-zwh3NnzMMgoVdJ5xUx - c
Cris140
08/22/2022, 6:08 PMWas it IPA? - c
Cris140
08/22/2022, 6:16 PMCan you pass me the model ID? I can test it here - u
{K EY1} (Kei)
08/22/2022, 6:24 PMSure, in a bit - u
{K EY1} (Kei)
08/22/2022, 6:24 PMYeah I can send a sample phrase in ipa if you'd like - u
{K EY1} (Kei)
08/22/2022, 7:48 PMhttps://drive.google.com/file/d/13u74DGxi4KNg_JGQ_2wbQDPsULTmOHLM/view?usp=sharing and you can try
for the input if you wantdiːs ɪst aɪnə lɪstə dɛɾ aʊfɡənɔmənən ʊnt fɛɾœfəntlɪçtən liːdɜ dɛɾ bɾiːtɪʃən rɔkbant moːtœɾheːɑːt. - c
Cris140
08/22/2022, 8:05 PMYep, I had no luck synthesizing with IPA - c
Cris140
08/22/2022, 8:06 PMThe problem is on the model though, not the system - u
{K EY1} (Kei)
08/22/2022, 10:53 PMI see... - u
{K EY1} (Kei)
08/22/2022, 10:53 PMIs there a chance it got corrupted? - u
{K EY1} (Kei)
08/22/2022, 10:54 PMShould i just retrain and see what happens? - c
Cris140
08/22/2022, 11:05 PMI don't think it got corrupted - c
Cris140
08/22/2022, 11:06 PMCause it wouldn't load if it were corrupted - c
Cris140
08/22/2022, 11:31 PMI'm using the pipeline to train IPA models now - c
Cris140
08/22/2022, 11:32 PMbut I think it would work just fine with the one you've sent - u
{K EY1} (Kei)
08/22/2022, 11:32 PMSo if it's not my method that's the problem, why isn't it synthesizing? - c
Cris140
08/22/2022, 11:32 PMDo you remember the training graphs?