ah okay

you can only train on one

Huh, welp, I guess it's time to experiment with the one I have... So this time, I'm gonna remove the normal input and just leave my Arpabet ones in place... And taking from your advice, I shouldn't overfit it... Alright, thanks!

Hmm

I'll submit an updated version of my Mei model probably sooner or later...

Help

Can't really help if you don't provide details

Something sounds odd about Hammond And me and Sid don't know why

(original audio)

Ah, easy fix

(synthesized)

Use the hifi notebook

1 sec

The what

@User https://colab.research.google.com/drive/1dxVcqe4m-AU8NAA1I1MW1N9HYBO_oii_?usp=sharing Paste your google share ID in tacotron override, turn off pronounciation_dictionary and run the cell

This basically boosts quallity, no need to know how it works.

Wait, do I retrain it?

Nah

phew.

Just use the existing model and paste it in the tacotron override and follow the instructions above I just send

Make sure to send some examples, really interested on how it impacts your model @User

Alright

comparisons

It sounds a bit better doesn't it?

A little bit, yeah.

Though idk what happened at the end

Maybe needs a bit more training, or just use shorter sentances.

oh no that just happends

just re try it

It always happens, but only on the HiFi