Training beginning now

Will check how the model turns out after a dozen or so epochs of training

--------------------------------------------------------------------------- ValueError Traceback (most recent call last) in () 6 7 #@markdown #### Execute this step to install Tacotron 2 and dependencies ----> 8 get_ipython().magic('tensorflow_version 1.x') 9 import os 10 get_ipython().system('git clone -q https://github.com/NVIDIA/tacotron2') 2 frames /usr/local/lib/python3.7/dist-packages/google/colab/_tensorflow_magics.py in _tensorflow_version(line) 39 40 Your notebook should be updated to use Tensorflow 2. ---> 41 See the guide at https://www.tensorflow.org/guide/migrate#migrate-from-tensorflow-1x-to-tensorflow-2.""" 42 )) 43 ValueError: Tensorflow 1 is unsupported in Colab. Your notebook should be updated to use Tensorflow 2. See the guide at https://www.tensorflow.org/guide/migrate#migrate-from-tensorflow-1x-to-tensorflow-2.

Is this Colab

Yeah

I made two models the other day using legacy and I didn’t have any problems after the first step

@Hange Zoë replace in the setup cell

%tensorflow_version 1.x

with

!pip install tensorflow==1.15

Did you add all katakana to the symbol set?

What symbol set?

Unlike my last attempt, I didn't write the g2p file

For this

The symbol set which is in symbols.py for legacy and idr where for pipeline

I tried twice but that didn't work

Not sure then I can check it out later

I'm making a model of Tamaki right now (it's gonna be for fakeyou for sure idk about uberduck) and it sounds like shit

"Haruhi help I'm in danger"

How do I get the embedding size of a model in Python?

i'm not sure of the details but check the model loading section of this notebook https://colab.research.google.com/drive/1wXWuhnw2pdfFy1L-pUzHfopW10W2GiJS look for

speaker_embedding

and replace it with

embedding

and it might work?

I will try it, I want to do a code that automatically chooses the symbol set based on the embedding size

ooooh

in pytorch you can do .shape

or is it .size

I have an error with auto transcribing...

By the way, this using CPU and not GPU

It worked! Thank you so much!

i was wondering if my dataset has calm voice and shouting voice

would that mess the model?

yes

either train 2 separate models or 2 speakers in 1 model

the uberduck pipeline emoji/emotion cant differ between 350 dataset of calm voice and 6 dataset of shouting voice?