Perfectpitcher left this server, and he no longer makes models for Uberduck.

Perfectpitcher88#5579 left this server

??

yeah i get the same

You mean the notebooks in #841437191073955920

Oh. That's news to me — I thought he had never joined this server.

Do not use that outdated notebook to train a Tacotron 2 model. Please use one of the correct notebooks in the #841437191073955920 channel instead.

so for the new notebook tensorflow 2 route notebook, I made sure the wav files are 16 bit pcm and the project is 22050 Hz, i previewed all the wav files on my computer and they are fine, but when i preview the audio tracks on the audio tensorboard, depending on which Mp3 file that are all re-encoded to 128 kbps that i sourced it from, some of the wav preview fragments in the tensorboard are fine while others have this long static after the audio clip that makes the wav length twice as long as it should be...i am not sure which step is causing the static while other wav file previews don't have it. not sure if it's just a tensorboard glitch or if some of my training wav data has something wrong with them...I can share the zip files but I remember doing this with the old notebook path last year with tensorflow 1 and it didn't take over 300 epochs but the new upgraded method with tensorflow 2 is taking more than twice as long with basically a few more mp3 source material

if the audio clip is short, does it automatically add extra static at the end to balance out the training data?

on batch sizes larger than 1

i got a problem

https://app.uberduck.ai/submit i'm getting this error when trying to submit to the website...is there anywhere else i could submit?

I'm afraid that's the only place where you can submit a model to Uberduck.

I did the same and it gave me the same error, even though I did exactly what the instruction said

No wonder why I thought using a pipeline notebook can be so hard...

I made a voice with 51 wavs and when it was done training, it says gibberish including the other voices I made (it even was 0.15)

hi...

Hi, I'm finishing up a dataset validator for Tacotron and other technologies based on the Lj Speech Dataset. The validator is based on plugins, so if anyone is interested, I can always add new plugins to the tool. The dataset with 19,361 entries is checked in less than 2 seconds. Currently, this tool has 8 plugins that check whether: - F001: all files in the wavs folder have been added to the transcription files - T001: transcription files exist (custom names can be provided as a parameter) and whether the wavs folder exists - T002: lines in the transcription files are not empty - T003: the files whose paths are added in the transcription files exist in the wavs folder - T004: transcriptions in the transcription files are not empty - T005: each transcription ends with one of the following characters: period, question mark or exclamation mark - T006: all lines have the same number of PIPE characters in the entire file - T007: duplicate paths to wav files are found in the transcription files The tool has been fairly tested, but if anyone is interested in testing it out, they can download it tomorrow (or the day after), after I push it to pypi. Anyone interested?

Hi, I try to run the Import libraries and define functions cell when it gives me this error: /content/tacotron2/text/__init__.py:74: SyntaxWarning: "is not" with a literal. Did you mean "!="? return s in _symbol_to_id and s is not '_' and s is not '~' /content/tacotron2/text/__init__.py:74: SyntaxWarning: "is not" with a literal. Did you mean "!="? return s in _symbol_to_id and s is not '_' and s is not '~' --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) in 396 397 # ---- DEFAULT PARAMETERS DEFINED HERE ---- --> 398 hparams = create_hparams() 399 model_filename = 'current_model' 400 hparams.training_files = "filelists/clipper_train_filelist.txt" /content/tacotron2/hparams.py in create_hparams(hparams_string, verbose) 6 """Create model hyperparameters. Parse nondefault from given string.""" 7 ----> 8 hparams = tf.contrib.training.HParams( 9 ################################ 10 # Experiment Parameters # AttributeError: module 'tensorflow' has no attribute 'contrib'

If you have trouble using any notebooks, make sure you use the correct notebooks, which are linked in the #notebooks channel, as those notebooks are the most up-to-date ones.

Can someone explain why everytime I make a voice, the voices just says gibberish all the time?

If they are pipeline models, you have to set ARPAbet to "True" when you send them to uberduck

No I mean on Tacotron.

and its not Pipeline

Dude, Uberduck Pipeline Tacotron 2 is an advanced version of Tacotron 2. That's what I mean by "pipeline".

Yea but we have to send csv files and I wont start a free trial

You don't need to start a free trail to submit a voice model to Uberduck

tried running the boring code in google collab's training notebook. gave me this

I checked stack overflow and it said that I needed to update my tensorflow version, but I'm pretty sure it's already on the latest version. the notebook with the upgrade script told me it was on 2.11

this is absolutely not normal behaviour, since the other cells don't run properly without this boring code operational