# tacotron-2-support
- u
{K EY1} (Kei)
07/17/2022, 1:51 AMOhh - u
(Dawn) Will Draw Fictional Women
07/17/2022, 2:01 AMyou can still train arpa on legacy - u
(Dawn) Will Draw Fictional Women
07/17/2022, 2:01 AMand you dont necessarily have to train arpa on pipeline - u
(Dawn) Will Draw Fictional Women
07/17/2022, 2:01 AMwe just dont have base models for it - u
{K EY1} (Kei)
07/17/2022, 2:01 AMunrelated but will a dataset with audio like this work in tacotron? i'm planning on using pipeline for this but if it would work better on legacy please tell me - u
(Dawn) Will Draw Fictional Women
07/17/2022, 2:01 AMyou can tryyy - u
(Dawn) Will Draw Fictional Women
07/17/2022, 2:01 AMbut go with crust - u
{K EY1} (Kei)
07/17/2022, 2:02 AMyeah i mainly meant like.. pipeline has arpa base models and also converts the text to arpabet doesnt it - u
(Dawn) Will Draw Fictional Women
07/17/2022, 2:02 AMaitch HATES effects - u
(Dawn) Will Draw Fictional Women
07/17/2022, 2:02 AMyes and optionally yes - u
{K EY1} (Kei)
07/17/2022, 2:02 AMoh ok! i'll try should i try with only a few wavs first to make sure it'll actually work before transcribing the whole 15 minute dataset? - u
(Dawn) Will Draw Fictional Women
07/17/2022, 2:03 AMmay as wll gun the whole thing down in one go - u
(Dawn) Will Draw Fictional Women
07/17/2022, 2:03 AMlow wav tests are frankly a terrible gague - u
{K EY1} (Kei)
07/17/2022, 2:03 AMah ok - u
{K EY1} (Kei)
07/17/2022, 2:05 AMsucks that all his voicelines are like that... a ton of the characters in hades have effects on their voices (ig cuz they're dead?) - u
{K EY1} (Kei)
07/17/2022, 2:05 AMi'll give it a shot though - z
Zzenith
07/18/2022, 12:37 AMIf anyone is trying to run the legacy notebook locally, one issue you'll run into is with
since tensorflow 2.0 doesn't support it. To save you some time if you're searching through this channel, just make it into an object instead with the values as properties. Here's an example, you'd just need to change the import statement of course. ill also note that this will impact the notebook as a whole when tensorflow 1.x gets completely deprecated august 1sttf.contrib.training.HParams - g
Gosmokeless28
07/18/2022, 12:44 AMTensorflow? In the legacy TT2 training notebook? - z
Zzenith
07/18/2022, 12:45 AMyes? - g
Gosmokeless28
07/18/2022, 12:45 AMI didn't know the legacy notebook had that - g
Gosmokeless28
07/18/2022, 12:45 AMI thought only the pipeline one did - z
Zzenith
07/18/2022, 12:46 AMya i haven't had the chance to mess with the pipeline one yet - z
Zzenith
07/18/2022, 12:47 AMim just really trying to run any of them locally to get away from colab gpu limits and stuff (and I know I should probably be using uberduck-ml-dev on github) but obviously they weren't meant to run outside of colab - c
Couch
07/18/2022, 12:47 AMmldev can absolutely run locally - z
Zzenith
07/18/2022, 12:47 AMyeah thats what I mean - z
Zzenith
07/18/2022, 12:48 AMi should probably be doing that since it's meant to, but I'm in deep on this one - z
Zzenith
07/18/2022, 12:48 AMsunk costs and all that - r
Reclezon
07/18/2022, 4:13 PMIf I'm already training a model, would randomly adding more data help it or should I restart? - h
hecko
07/18/2022, 6:33 PMit will help but you might be best off restarting - z
Zzenith
07/18/2022, 7:28 PMit will only help to a limited extent, so by restarting you can get better results