-
Notifications
You must be signed in to change notification settings - Fork 9
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
1. add use_text_encoder into config; 2. remove text_encoder in grad_t…
…ts and diffwave
- Loading branch information
1 parent
6a131e7
commit dceaf0f
Showing
15 changed files
with
324 additions
and
59 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,76 @@ | ||
# experiment | ||
dataset: vctk | ||
train_meta: data/vctk/train_nodev_all/metadata.csv | ||
dev_meta: data/vctk/dev_all/metadata.csv | ||
train_set: train_nodev_all | ||
dev_set: dev_all | ||
|
||
|
||
# encoder-decoder | ||
ling_enc: vqwav2vec | ||
spk_enc: utt_dvec | ||
pros_enc: ppgvc_f0 | ||
decoder: DiffWave | ||
mel_type: vits_spec # will not use it | ||
pros_stats: dump/vctk/train_nodev_all/fastspeech2_pitch_energy/train_nodev_all.npy | ||
|
||
# training | ||
fp16_run: !!bool False | ||
epochs: 2000 | ||
save_freq: 1 # save ckpt frequency | ||
show_freq: 100 # show training information frequency | ||
load_only_params: !!bool False | ||
seed: !!int 1234 | ||
trainer: DiffWaveTrainer | ||
ngpu: 1 | ||
|
||
#dataloader | ||
dataset_class: VITSDataset | ||
sampling_rate: !!int 24000 | ||
vits_hop_size: !!int 240 | ||
spec_max_len: !!int 60 | ||
sort: !!bool False | ||
dump_dir: dump | ||
num_workers: !!int 8 | ||
batch_size: 16 | ||
drop_last: !!bool True | ||
rm_long_utt: !!bool False # remove too long utterances from metadata | ||
max_utt_duration: !!float 10.0 # max utterance duration (seconds) | ||
|
||
|
||
# decoder params | ||
decoder_params: | ||
segment_size: 14400 | ||
use_text_encoder: !!bool False | ||
input_dim: !!int 512 | ||
spk_emb_dim: !!int 256 | ||
prosodic_rep_type: continuous | ||
prosodic_net: | ||
hidden_dim: 64 | ||
inter_channels: !!int 64 | ||
hidden_channels: !!int 64 | ||
filter_channels: !!int 128 | ||
n_heads: !!int 2 | ||
n_layers: !!int 6 | ||
kernel_size: !!int 3 | ||
p_dropout: !!float 0.1 | ||
residual_channels: 64 | ||
residual_layers: 30 | ||
dilation_cycle_length: 10 | ||
|
||
|
||
|
||
#optimizer & scheduler | ||
optimizer: | ||
lr: !!float 2e-4 | ||
|
||
# loss hyper-parameters | ||
losses: | ||
alpha: 1. | ||
|
||
|
||
|
||
|
||
|
||
|
||
|
76 changes: 76 additions & 0 deletions
76
configs/vctk_vqwav2vec_uttdvec_ppgvcf0_gradtts_bigvgan.yaml
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,76 @@ | ||
# experiment | ||
dataset: vctk | ||
train_meta: data/vctk/train_nodev_all/metadata.csv | ||
dev_meta: data/vctk/dev_all/metadata.csv | ||
train_set: train_nodev_all | ||
dev_set: dev_all | ||
|
||
|
||
# encoder-decoder | ||
ling_enc: vqwav2vec | ||
spk_enc: utt_dvec | ||
pros_enc: ppgvc_f0 | ||
decoder: GradTTS | ||
mel_type: norm_bigvgan_mel | ||
vocoder: bigvgan | ||
mel_stats: dump/vctk/train_nodev_all/bigvgan_mel/train_nodev_all.npy | ||
|
||
# training | ||
fp16_run: !!bool False | ||
epochs: 2000 | ||
save_freq: 1 # save ckpt frequency | ||
show_freq: 100 # show training information frequency | ||
load_only_params: !!bool False | ||
seed: !!int 1234 | ||
trainer: GradTTSTrainer | ||
ngpu: 1 | ||
|
||
#dataloader | ||
dataset_class: Dataset | ||
mel_segment_length: !!int 128 | ||
sort: !!bool False | ||
dump_dir: dump | ||
num_workers: !!int 8 | ||
batch_size: 16 | ||
drop_last: !!bool True | ||
rm_long_utt: !!bool True # remove too long utterances from metadata | ||
max_utt_duration: !!float 10.0 # max utterance duration (seconds) | ||
|
||
|
||
# decoder params | ||
decoder_params: | ||
use_text_encoder: !!bool False | ||
use_prior_loss: !!bool False | ||
n_feats: !!int 100 | ||
input_dim: !!int 512 | ||
spk_emb_dim: !!int 256 | ||
prosodic_rep_type: continuous | ||
prosodic_net: | ||
hidden_dim: !!int 100 | ||
n_enc_channels: !!int 192 | ||
filter_channels: !!int 768 | ||
filter_channels_dp: !!int 256 | ||
n_enc_layers: !!int 6 | ||
enc_kernel: !!int 3 | ||
enc_dropout: !!float 0.1 | ||
n_heads: !!int 2 | ||
window_size: !!int 4 | ||
dec_dim: !!int 64 | ||
beta_min: !!float 0.05 | ||
beta_max: !!float 20.0 | ||
pe_scale: !!int 1000 # 1 for `grad-tts-old.pt` checkpoint | ||
|
||
#optimizer & scheduler | ||
optimizer: | ||
lr: !!float 1e-4 | ||
|
||
# loss hyper-parameters | ||
losses: | ||
alpha: 1. | ||
|
||
|
||
|
||
|
||
|
||
|
||
|
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
79 changes: 79 additions & 0 deletions
79
configs/vctk_vqwav2vec_uttdvec_ppgvcf0_tacoar_bigvgan.yaml
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,79 @@ | ||
# experiment | ||
dataset: vctk | ||
train_meta: data/vctk/train_nodev_all/metadata.csv | ||
dev_meta: data/vctk/dev_all/metadata.csv | ||
train_set: train_nodev_all | ||
dev_set: dev_all | ||
|
||
|
||
# encoder-decoder | ||
ling_enc: vqwav2vec | ||
spk_enc: utt_dvec | ||
pros_enc: ppgvc_f0 | ||
decoder: TacoAR | ||
mel_type: norm_bigvgan_mel | ||
vocoder: bigvgan | ||
mel_stats: dump/vctk/train_nodev_all/bigvgan_mel/train_nodev_all.npy | ||
|
||
|
||
# training | ||
fp16_run: !!bool True | ||
epochs: 200 | ||
save_freq: 2 # save ckpt frequency | ||
show_freq: 10 | ||
load_only_params: !!bool False | ||
seed: !!int 1234 | ||
trainer: TacoARTrainer | ||
ngpu: 1 | ||
|
||
#dataloader | ||
dataset_class: Dataset | ||
sort: !!bool True | ||
dump_dir: dump | ||
num_workers: !!int 8 | ||
batch_size: 32 | ||
drop_last: !!bool True | ||
rm_long_utt: !!bool True # remove too long utterances from metadata | ||
max_utt_duration: !!float 10.0 # max utterance duration (seconds) | ||
|
||
|
||
# decoder params | ||
decoder_params: | ||
prosodic_rep_type: continuous | ||
prosodic_net: | ||
hidden_dim: 1024 | ||
input_dim: 512 | ||
output_dim: 100 | ||
resample_ratio: 1 | ||
spk_emb_integration_type: concat # add or concat | ||
spk_emb_dim: 256 | ||
ar: True | ||
encoder_type: "taco2" | ||
hidden_dim: 1024 | ||
prenet_layers: 2 # if set 0, no prenet is used | ||
prenet_dim: 256 | ||
prenet_dropout_rate: 0.5 | ||
lstmp_layers: 2 | ||
lstmp_dropout_rate: 0.2 | ||
lstmp_proj_dim: 256 | ||
lstmp_layernorm: False | ||
|
||
#optimizer & scheduler | ||
optimizer: | ||
weight_decay: 0.0 | ||
betas: [0.9,0.99] | ||
lr: !!float 1e-4 | ||
scheduler: | ||
num_training_steps: 500000 | ||
num_warmup_steps: 4000 | ||
|
||
# loss hyper-parameters | ||
losses: | ||
alpha: 1. | ||
|
||
|
||
|
||
|
||
|
||
|
||
|
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.