Medix-AI/coqui-tts
Medix-AI
/
coqui-tts
mirror of https://github.com/coqui-ai/TTS.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

feat: updated recipes and lr fix (#1718) 

- updated the recipes activating more losses for more stable training
- re-enabling guided attention loss
- fixed a bug about not the correct lr fetched for logging
This commit is contained in:
a-froghyar 2022-07-12 15:00:53 +02:00 committed by GitHub
parent 48a4f3647f
commit 34b80e0280
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 12 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
TTS/utils/capacitron_optimizer.py
View File

@ -34,6 +34,8 @@ class CapacitronOptimizer:
self.primary_optimizer.zero_grad() self.primary_optimizer.zero_grad()
def step(self): def step(self):
# Update param groups to display the correct learning rate
self.param_groups = self.primary_optimizer.param_groups
self.primary_optimizer.step() self.primary_optimizer.step()
def zero_grad(self): def zero_grad(self):

12
recipes/blizzard2013/tacotron1-Capacitron/train_capacitron_t1.py
View File

@ -48,7 +48,6 @@ config = TacotronConfig(
precompute_num_workers=24, precompute_num_workers=24,
run_eval=True, run_eval=True,
test_delay_epochs=5, test_delay_epochs=5,
ga_alpha=0.0,
r=2, r=2,
optimizer="CapacitronOptimizer", optimizer="CapacitronOptimizer",
optimizer_params={"RAdam": {"betas": [0.9, 0.998], "weight_decay": 1e-6}, "SGD": {"lr": 1e-5, "momentum": 0.9}}, optimizer_params={"RAdam": {"betas": [0.9, 0.998], "weight_decay": 1e-6}, "SGD": {"lr": 1e-5, "momentum": 0.9}},
@ -68,16 +67,15 @@ config = TacotronConfig(
datasets=[dataset_config], datasets=[dataset_config],
lr=1e-3, lr=1e-3,
lr_scheduler="StepwiseGradualLR", lr_scheduler="StepwiseGradualLR",
lr_scheduler_params={"gradual_learning_rates": [[0, 1e-3], [2e4, 5e-4], [4e5, 3e-4], [6e4, 1e-4], [8e4, 5e-5]]}, lr_scheduler_params={"gradual_learning_rates": [[0, 1e-3], [2e4, 5e-4], [4e4, 3e-4], [6e4, 1e-4], [8e4, 5e-5]]},
scheduler_after_epoch=False, # scheduler doesn't work without this flag scheduler_after_epoch=False, # scheduler doesn't work without this flag
# Need to experiment with these below for capacitron
loss_masking=False, loss_masking=False,
decoder_loss_alpha=1.0, decoder_loss_alpha=1.0,
postnet_loss_alpha=1.0, postnet_loss_alpha=1.0,
postnet_diff_spec_alpha=0.0, postnet_diff_spec_alpha=1.0,
decoder_diff_spec_alpha=0.0, decoder_diff_spec_alpha=1.0,
decoder_ssim_alpha=0.0, decoder_ssim_alpha=1.0,
postnet_ssim_alpha=0.0, postnet_ssim_alpha=1.0,
) )
ap = AudioProcessor(**config.audio.to_dict()) ap = AudioProcessor(**config.audio.to_dict())

12
recipes/blizzard2013/tacotron2-Capacitron/train_capacitron_t2.py
View File

@ -52,7 +52,6 @@ config = Tacotron2Config(
precompute_num_workers=24, precompute_num_workers=24,
run_eval=True, run_eval=True,
test_delay_epochs=5, test_delay_epochs=5,
ga_alpha=0.0,
r=2, r=2,
optimizer="CapacitronOptimizer", optimizer="CapacitronOptimizer",
optimizer_params={"RAdam": {"betas": [0.9, 0.998], "weight_decay": 1e-6}, "SGD": {"lr": 1e-5, "momentum": 0.9}}, optimizer_params={"RAdam": {"betas": [0.9, 0.998], "weight_decay": 1e-6}, "SGD": {"lr": 1e-5, "momentum": 0.9}},
@ -77,21 +76,20 @@ config = Tacotron2Config(
"gradual_learning_rates": [ "gradual_learning_rates": [
[0, 1e-3], [0, 1e-3],
[2e4, 5e-4], [2e4, 5e-4],
[4e5, 3e-4], [4e4, 3e-4],
[6e4, 1e-4], [6e4, 1e-4],
[8e4, 5e-5], [8e4, 5e-5],
] ]
}, },
scheduler_after_epoch=False, # scheduler doesn't work without this flag scheduler_after_epoch=False, # scheduler doesn't work without this flag
seq_len_norm=True, seq_len_norm=True,
# Need to experiment with these below for capacitron
loss_masking=False, loss_masking=False,
decoder_loss_alpha=1.0, decoder_loss_alpha=1.0,
postnet_loss_alpha=1.0, postnet_loss_alpha=1.0,
postnet_diff_spec_alpha=0.0, postnet_diff_spec_alpha=1.0,
decoder_diff_spec_alpha=0.0, decoder_diff_spec_alpha=1.0,
decoder_ssim_alpha=0.0, decoder_ssim_alpha=1.0,
postnet_ssim_alpha=0.0, postnet_ssim_alpha=1.0,
) )
ap = AudioProcessor(**config.audio.to_dict()) ap = AudioProcessor(**config.audio.to_dict())