Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 27, 2019

• Merge pull request fatchord#156 from sih4sing5hong5/patch-1

  Show description for 3595219
  fatchord

  authored
  3595219

  Copy full SHA for 3595219

Commits on Nov 23, 2019

• [README] fix quote

  sih4sing5hong5

  authored
  7a65481

  Copy full SHA for 7a65481

Commits on Nov 7, 2019

• Fix batched generation in quick_start.py

  fatchord

  committed
  521859d

  Copy full SHA for 521859d

• Fix crossfade misalignment

  fatchord

  committed
  544cd5d

  Copy full SHA for 544cd5d

Commits on Sep 8, 2019

• Merge pull request fatchord#126 from TheButlah/master

  Show description for 12922a7
  fatchord

  authored
  12922a7

  Copy full SHA for 12922a7

Commits on Sep 5, 2019

• Made r value in quick_start.py for tts_model no longer hardcoded

  TheButlah

  committed
  31179a6

  Copy full SHA for 31179a6

• Fix quick_start.py

  fatchord

  committed
  fb8fcb4

  Copy full SHA for fb8fcb4

• Generalise checkpointing for both model types

  fatchord

  committed
  72d28b0

  Copy full SHA for 72d28b0

Commits on Aug 22, 2019

• Allowed gen_wavernn.py to accept mel spectrogram .npy

  TheButlah

  committed
  942dddb

  Copy full SHA for 942dddb

Commits on Aug 14, 2019

• Updated readme to fix gen_tacotron commands

  TheButlah

  committed
  1798680

  Copy full SHA for 1798680

• Made vocoder and synth weights explicitly separated for inference

  TheButlah

  committed
  352c51a

  Copy full SHA for 352c51a

Commits on Aug 13, 2019

• Made r=2 in last session in tts hparams schedule

  TheButlah

  committed
  d2f0bb9

  Copy full SHA for d2f0bb9

• Made epochs 1-indexed

  TheButlah

  committed
  6b79dbb

  Copy full SHA for 6b79dbb

Commits on Aug 12, 2019

• Fixed bug where current_step in training loop not updated

  TheButlah

  committed
  4e940e1

  Copy full SHA for 4e940e1

Commits on Aug 10, 2019

• Made force_train in tacotron actually do something

  TheButlah

  committed
  1b4aaa7

  Copy full SHA for 1b4aaa7

• Added tts_stop_threshold hparam

  TheButlah

  committed
  f62bbc5

  Copy full SHA for f62bbc5

• Fix path name issue for attention plots

  TheButlah

  committed
  8652895

  Copy full SHA for 8652895

• Finished adding Griffin-Lim vocoder, backward compat with saved models

  TheButlah

  committed
  b43cced

  Copy full SHA for b43cced

Commits on Aug 9, 2019

• Began adding Griffin-Lim vocoder

  TheButlah

  committed
  d0cc6b4

  Copy full SHA for d0cc6b4

Commits on Aug 7, 2019

• Added note to call model.to(device) after restore_checkpoint()

  TheButlah

  committed
  d2fc0a5

  Copy full SHA for d2fc0a5

• Made inference scripts use Pathlib, fixes minor bug

  TheButlah

  committed
  d30754f

  Copy full SHA for d30754f

• Fixed train_wavernn and inference scripts to use new ckpt code

  TheButlah

  committed
  e185a96

  Copy full SHA for e185a96

Commits on Jul 30, 2019

• Added hparams pseudo-module to enable changing hparams at runtime

  TheButlah

  committed
  697ec9c

  Copy full SHA for 697ec9c

• Fixed issue where preprocess args were referencing hp before defined

  TheButlah

  committed
  eaa0b25

  Copy full SHA for eaa0b25

• Removed single letter flags and added error message for import from path

  

  TheButlah

  authored and

  Ryan Butler

  committed
  f71131d

  Copy full SHA for f71131d

• Made preprocess.py accept argument to control location of hparams.py

  

  TheButlah

  authored and

  Ryan Butler

  committed
  146cfe1

  Copy full SHA for 146cfe1

• Enabled specifying hparams location in train_tacotron/wavernn

  

  TheButlah

  authored and

  Ryan Butler

  committed
  45707ef

  Copy full SHA for 45707ef

Commits on Jul 29, 2019

• Fixed bug in create_gta_features where enumeated over int

  TheButlah

  committed
  54bb541

  Copy full SHA for 54bb541

• Silenced the printing of when checkpoints are saved during training

  TheButlah

  committed
  c549267

  Copy full SHA for c549267

Commits on Jul 26, 2019

• Fixed bug where paths was not in scope for taco train loop

  TheButlah

  committed
  212b8b4

  Copy full SHA for 212b8b4

• Fixed train_tacotron.py so that it works with new way of checkpointing

  TheButlah

  committed
  924fbeb

  Copy full SHA for 924fbeb

• Added save_checkpoint and restore_checkpoint functions

  Show description for c738ee5
  TheButlah

  committed
  c738ee5

  Copy full SHA for c738ee5

• Made models use the regular way of detecting training vs eval

  TheButlah

  committed
  90bb56d

  Copy full SHA for 90bb56d

• Fixed issue where attention tensor used as numpy array

  TheButlah

  committed
  62abd12

  Copy full SHA for 62abd12

• Fixed missing import and using x11

  Show description for 1ed4591
  TheButlah

  committed
  1ed4591

  Copy full SHA for 1ed4591

Pagination

Previous

Next