You can use mine

It's in Portuguese, but it's almost the same process as the original

I see, thanks for sharing Let's see if I manage to get the model working (finally)

Just run the cells, put the model id and hifigan id, change the symbol set, and you're ready to go

It's working yahooo I'll pin it on my drive cus I'll be continuing the work on this model in the future

Nice! Can you share a sample?

I could buttt it's severely undertrained

It didn't save the 40th epoch model for some reason

I just have Tensorboard outputs...

It's okay tho because I needed to re-edit the audio nonetheless

I was about to send an example when inferencing I realised the second batch of data I added in today had the transcriptions offset by one...

Will have to re-train tomorrow, again

Do we know if it's possible to continue training a model from a previous session?

May want to fine-tune mine a little more

yes its possible

I'm guessing one needs to set the model name the same as from the previous session and it will resume training from there

yes

and same symbol set and all that

Alrighty

The graph looks OK

but the audio doesn't seem to be

Oh it's because Tensorboard uses a fast vocoder for preview rendering

It sounds just fine in the synthesis notebook

help please

The colab notebook for testing a voice model that's listed here in the "Uberduck Written Tutorial" is broken because it uses Tensorflow v1 (which Google Colab no longer supports): https://uberduck-guide.gitbook.io/uberduck-written-tutorial/testing-audio-model/hifi-gan Does anyone have a Google Colab notebook I could use to test a model created by following the Colab notebook that's listed on the Uberduck Written Tutorials' page for "Simplified Training" that uses Tensorflow version 2 please? https://uberduck-guide.gitbook.io/uberduck-written-tutorial/training-time/training-audio-with-tacotron-2-colab-page/simplified-training-page Here's the error message I get from running the testing-a-voice-model notebook: --------------------------------------------------------------------------- NameError Traceback (most recent call last) in 91 try: ---> 92 initialized 93 except NameError: NameError: name 'initialized' is not defined During handling of the above exception, another exception occurred: ValueError Traceback (most recent call last) 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. ------------------------------------------------------------ Thanks!

How are you supposed to load in a model from Google Drive to the second notebook listed in #841437191073955920, the one for testing a model? The picture shows my configs - tacotron_id contains the ID of the Google Drive object of the model I'm trying to synthesize .wavs from. However, the .wav outputted when the notebook is run with these configs is spoken by the voice of the woman from the LJSpeech dataset, rather than the voice I trained into my model. Any help is appreciated, thanks!

found my answer in the guide, looks like I didn't use enough wavs to train it: https://uberduck-guide.gitbook.io/uberduck-written-tutorial/troubleshoot-and-bugs/troubleshoot-and-bugs#i-got-this-ladys-default-voice-which-is-thats-not-my-model.-what-to-do

You can use smaller amounts of wavs, but you need to make sure the batch size is lower than the amount of wavs you have

training on this notebook https://colab.research.google.com/drive/1jF-Otw2_ssEcus4ISaIZu3QDmtifUvyY?authuser=1#scrollTo=5c6sPMK4yhe0

please ping me if youve got an answer