chore: import upstream snapshot with attribution
This commit is contained in:
wehub-resource-sync
@@ -0,0 +1,144 @@
|
||||
base_config:
|
||||
- configs/base.yaml
|
||||
|
||||
task_cls: training.acoustic_task.AcousticTask
|
||||
|
||||
dictionaries: {}
|
||||
extra_phonemes: []
|
||||
merged_phoneme_groups: []
|
||||
datasets: []
|
||||
|
||||
vocoder: NsfHifiGAN
|
||||
vocoder_ckpt: checkpoints/pc_nsf_hifigan_44.1k_hop512_128bin_2025.02/model.ckpt
|
||||
audio_sample_rate: 44100
|
||||
audio_num_mel_bins: 128
|
||||
hop_size: 512 # Hop size.
|
||||
fft_size: 2048 # FFT size.
|
||||
win_size: 2048 # FFT size.
|
||||
fmin: 40
|
||||
fmax: 16000
|
||||
|
||||
binarization_args:
|
||||
shuffle: true
|
||||
num_workers: 0
|
||||
augmentation_args:
|
||||
random_pitch_shifting:
|
||||
enabled: false
|
||||
range: [-5., 5.]
|
||||
scale: 0.75
|
||||
fixed_pitch_shifting:
|
||||
enabled: false
|
||||
targets: [-5., 5.]
|
||||
scale: 0.5
|
||||
random_time_stretching:
|
||||
enabled: false
|
||||
range: [0.5, 2.]
|
||||
scale: 0.75
|
||||
|
||||
binary_data_dir: 'data/opencpop/binary'
|
||||
binarizer_cls: preprocessing.acoustic_binarizer.AcousticBinarizer
|
||||
spec_min: [-12]
|
||||
spec_max: [0]
|
||||
mel_vmin: -14.
|
||||
mel_vmax: 4.
|
||||
mel_base: 'e'
|
||||
energy_smooth_width: 0.06
|
||||
breathiness_smooth_width: 0.06
|
||||
voicing_smooth_width: 0.06
|
||||
tension_smooth_width: 0.06
|
||||
|
||||
use_lang_id: false
|
||||
num_lang: 1
|
||||
use_spk_id: false
|
||||
num_spk: 1
|
||||
use_mix_ln: false
|
||||
mix_ln_layer: [0, 2]
|
||||
use_energy_embed: false
|
||||
use_breathiness_embed: false
|
||||
use_voicing_embed: false
|
||||
use_tension_embed: false
|
||||
use_key_shift_embed: false
|
||||
use_speed_embed: false
|
||||
|
||||
diffusion_type: reflow
|
||||
time_scale_factor: 1000
|
||||
timesteps: 1000
|
||||
max_beta: 0.02
|
||||
enc_ffn_kernel_size: 3
|
||||
use_rope: true
|
||||
rope_interleaved: false
|
||||
use_stretch_embed: true
|
||||
use_variance_scaling: true
|
||||
rel_pos: true
|
||||
sampling_algorithm: euler
|
||||
sampling_steps: 20
|
||||
diff_accelerator: ddim
|
||||
diff_speedup: 10
|
||||
hidden_size: 384
|
||||
backbone_type: 'lynxnet2'
|
||||
backbone_args:
|
||||
num_channels: 1024
|
||||
num_layers: 6
|
||||
kernel_size: 31
|
||||
dropout_rate: 0.0
|
||||
use_conditioner_cache: true
|
||||
glu_type: 'atanglu'
|
||||
main_loss_type: l2
|
||||
main_loss_log_norm: false
|
||||
schedule_type: 'linear'
|
||||
|
||||
# shallow diffusion
|
||||
use_shallow_diffusion: true
|
||||
T_start: 0.4
|
||||
T_start_infer: 0.4
|
||||
K_step: 400
|
||||
K_step_infer: 400
|
||||
|
||||
shallow_diffusion_args:
|
||||
train_aux_decoder: true
|
||||
train_diffusion: true
|
||||
val_gt_start: false
|
||||
aux_decoder_arch: convnext
|
||||
aux_decoder_args:
|
||||
num_channels: 512
|
||||
num_layers: 6
|
||||
kernel_size: 7
|
||||
dropout_rate: 0.1
|
||||
aux_decoder_grad: 0.1
|
||||
|
||||
lambda_aux_mel_loss: 0.2
|
||||
|
||||
# train and eval
|
||||
num_sanity_val_steps: 1
|
||||
optimizer_args:
|
||||
optimizer_cls: modules.optimizer.muon.Muon_AdamW
|
||||
lr: 0.0006
|
||||
muon_args:
|
||||
weight_decay: 0.1
|
||||
adamw_args:
|
||||
weight_decay: 0.0
|
||||
lr_scheduler_args:
|
||||
step_size: 5000
|
||||
gamma: 0.8
|
||||
max_batch_frames: 50000
|
||||
max_batch_size: 64
|
||||
dataset_size_key: 'lengths'
|
||||
val_with_vocoder: true
|
||||
val_check_interval: 4000
|
||||
num_valid_plots: 10
|
||||
max_updates: 100000
|
||||
num_ckpt_keep: 8
|
||||
permanent_ckpt_start: 60000
|
||||
permanent_ckpt_interval: 10000
|
||||
|
||||
finetune_enabled: false
|
||||
finetune_ckpt_path: null
|
||||
|
||||
finetune_ignored_params:
|
||||
- model.fs2.encoder.embed_tokens
|
||||
- model.fs2.txt_embed
|
||||
- model.fs2.spk_embed
|
||||
finetune_strict_shapes: true
|
||||
|
||||
freezing_enabled: false
|
||||
frozen_params: []
|
||||
@@ -0,0 +1,94 @@
|
||||
# task
|
||||
task_cls: null
|
||||
|
||||
#############
|
||||
# dataset
|
||||
#############
|
||||
sort_by_len: true
|
||||
datasets: []
|
||||
binary_data_dir: null
|
||||
binarizer_cls: null
|
||||
binarization_args:
|
||||
shuffle: false
|
||||
num_workers: 0
|
||||
|
||||
audio_sample_rate: 44100
|
||||
hop_size: 512
|
||||
win_size: 2048
|
||||
fft_size: 2048 # Extra window size is filled with 0 paddings to match this parameter
|
||||
sampler_frame_count_grid: 6
|
||||
ds_workers: 4
|
||||
dataloader_prefetch_factor: 2
|
||||
|
||||
#########
|
||||
# model
|
||||
#########
|
||||
hidden_size: 384
|
||||
dropout: 0.1
|
||||
use_pos_embed: true
|
||||
enc_layers: 4
|
||||
num_heads: 2
|
||||
enc_ffn_kernel_size: 3
|
||||
ffn_act: gelu
|
||||
use_spk_id: false
|
||||
|
||||
###########
|
||||
# optimization
|
||||
###########
|
||||
optimizer_args:
|
||||
optimizer_cls: modules.optimizer.muon.Muon_AdamW
|
||||
lr: 0.0006
|
||||
beta1: 0.9
|
||||
beta2: 0.98
|
||||
weight_decay: 0
|
||||
lr_scheduler_args:
|
||||
scheduler_cls: torch.optim.lr_scheduler.StepLR
|
||||
step_size: 5000
|
||||
gamma: 0.8
|
||||
clip_grad_norm: 1
|
||||
|
||||
###########
|
||||
# train and eval
|
||||
###########
|
||||
num_ckpt_keep: 5
|
||||
accumulate_grad_batches: 1
|
||||
log_interval: 100
|
||||
num_sanity_val_steps: 1 # steps of validation at the beginning
|
||||
val_check_interval: 4000
|
||||
max_updates: 100000
|
||||
max_batch_frames: 50000
|
||||
max_batch_size: 100000
|
||||
max_val_batch_frames: 60000
|
||||
max_val_batch_size: 1
|
||||
pe: parselmouth
|
||||
pe_ckpt: 'checkpoints/rmvpe/model.pt'
|
||||
hnsep: vr
|
||||
hnsep_ckpt: 'checkpoints/vr/model.pt'
|
||||
f0_min: 65
|
||||
f0_max: 1100
|
||||
num_valid_plots: 10
|
||||
|
||||
###########
|
||||
# pytorch lightning
|
||||
# Read https://lightning.ai/docs/pytorch/stable/common/trainer.html#trainer-class-api for possible values
|
||||
###########
|
||||
pl_trainer_accelerator: 'auto'
|
||||
pl_trainer_devices: 'auto'
|
||||
pl_trainer_precision: '16-mixed'
|
||||
pl_trainer_num_nodes: 1
|
||||
pl_trainer_strategy:
|
||||
name: auto
|
||||
process_group_backend: nccl
|
||||
find_unused_parameters: false
|
||||
nccl_p2p: true
|
||||
|
||||
###########
|
||||
# finetune
|
||||
###########
|
||||
finetune_enabled: false
|
||||
finetune_ckpt_path: null
|
||||
finetune_ignored_params: []
|
||||
finetune_strict_shapes: true
|
||||
|
||||
freezing_enabled: false
|
||||
frozen_params: []
|
||||
@@ -0,0 +1,128 @@
|
||||
base_config:
|
||||
- configs/acoustic.yaml
|
||||
|
||||
dictionaries:
|
||||
zh: dictionaries/opencpop-extension.txt
|
||||
extra_phonemes: []
|
||||
merged_phoneme_groups: []
|
||||
|
||||
datasets:
|
||||
- raw_data_dir: data/xxx1/raw
|
||||
speaker: speaker1
|
||||
spk_id: 0
|
||||
language: zh
|
||||
test_prefixes:
|
||||
- wav1
|
||||
- wav2
|
||||
- wav3
|
||||
- wav4
|
||||
- wav5
|
||||
- raw_data_dir: data/xxx2/raw
|
||||
speaker: speaker2
|
||||
spk_id: 1
|
||||
language: zh
|
||||
test_prefixes:
|
||||
- wav1
|
||||
- wav2
|
||||
- wav3
|
||||
- wav4
|
||||
- wav5
|
||||
|
||||
binary_data_dir: data/xxx/binary
|
||||
binarization_args:
|
||||
num_workers: 0
|
||||
pe: parselmouth
|
||||
pe_ckpt: 'checkpoints/rmvpe/model.pt'
|
||||
hnsep: vr
|
||||
hnsep_ckpt: 'checkpoints/vr/model.pt'
|
||||
vocoder: NsfHifiGAN
|
||||
vocoder_ckpt: checkpoints/nsf_hifigan_44.1k_hop512_128bin_2024.02/model.ckpt
|
||||
|
||||
use_lang_id: false
|
||||
num_lang: 1
|
||||
use_spk_id: false
|
||||
num_spk: 1
|
||||
|
||||
use_mix_ln: false
|
||||
mix_ln_layer: [0, 2]
|
||||
|
||||
# NOTICE: before enabling variance embeddings, please read the docs at
|
||||
# https://github.com/openvpi/DiffSinger/tree/main/docs/BestPractices.md#choosing-variance-parameters
|
||||
use_energy_embed: false
|
||||
use_breathiness_embed: false
|
||||
use_voicing_embed: false
|
||||
use_tension_embed: false
|
||||
|
||||
use_key_shift_embed: true
|
||||
use_speed_embed: true
|
||||
|
||||
augmentation_args:
|
||||
random_pitch_shifting:
|
||||
enabled: true
|
||||
range: [-5., 5.]
|
||||
scale: 0.75
|
||||
fixed_pitch_shifting:
|
||||
enabled: false
|
||||
targets: [-5., 5.]
|
||||
scale: 0.5
|
||||
random_time_stretching:
|
||||
enabled: true
|
||||
range: [0.5, 2.]
|
||||
scale: 0.75
|
||||
|
||||
# diffusion and shallow diffusion
|
||||
diffusion_type: reflow
|
||||
enc_ffn_kernel_size: 3
|
||||
use_rope: true
|
||||
rope_interleaved: false
|
||||
use_stretch_embed: true
|
||||
use_variance_scaling: true
|
||||
use_shallow_diffusion: true
|
||||
T_start: 0.4
|
||||
T_start_infer: 0.4
|
||||
K_step: 300
|
||||
K_step_infer: 300
|
||||
hidden_size: 384
|
||||
backbone_type: 'lynxnet2'
|
||||
backbone_args:
|
||||
num_channels: 1024
|
||||
num_layers: 6
|
||||
kernel_size: 31
|
||||
dropout_rate: 0.0
|
||||
use_conditioner_cache: true
|
||||
glu_type: 'atanglu'
|
||||
shallow_diffusion_args:
|
||||
train_aux_decoder: true
|
||||
train_diffusion: true
|
||||
val_gt_start: false
|
||||
aux_decoder_arch: convnext
|
||||
aux_decoder_args:
|
||||
num_channels: 512
|
||||
num_layers: 6
|
||||
kernel_size: 7
|
||||
dropout_rate: 0.1
|
||||
aux_decoder_grad: 0.1
|
||||
lambda_aux_mel_loss: 0.2
|
||||
|
||||
optimizer_args:
|
||||
optimizer_cls: modules.optimizer.muon.Muon_AdamW
|
||||
lr: 0.0006
|
||||
muon_args:
|
||||
weight_decay: 0.1
|
||||
adamw_args:
|
||||
weight_decay: 0.0
|
||||
lr_scheduler_args:
|
||||
step_size: 5000
|
||||
gamma: 0.8
|
||||
max_batch_frames: 50000
|
||||
max_batch_size: 64
|
||||
max_updates: 100000
|
||||
|
||||
num_valid_plots: 10
|
||||
val_with_vocoder: true
|
||||
val_check_interval: 4000
|
||||
num_ckpt_keep: 8
|
||||
permanent_ckpt_start: 60000
|
||||
permanent_ckpt_interval: 10000
|
||||
pl_trainer_devices: 'auto'
|
||||
pl_trainer_precision: '16-mixed'
|
||||
@@ -0,0 +1,139 @@
|
||||
base_config:
|
||||
- configs/variance.yaml
|
||||
|
||||
dictionaries:
|
||||
zh: dictionaries/opencpop-extension.txt
|
||||
extra_phonemes: []
|
||||
merged_phoneme_groups: []
|
||||
|
||||
datasets:
|
||||
- raw_data_dir: data/xxx1/raw
|
||||
speaker: speaker1
|
||||
spk_id: 0
|
||||
language: zh
|
||||
test_prefixes:
|
||||
- wav1
|
||||
- wav2
|
||||
- wav3
|
||||
- wav4
|
||||
- wav5
|
||||
- raw_data_dir: data/xxx2/raw
|
||||
speaker: speaker2
|
||||
spk_id: 1
|
||||
language: zh
|
||||
test_prefixes:
|
||||
- wav1
|
||||
- wav2
|
||||
- wav3
|
||||
- wav4
|
||||
- wav5
|
||||
|
||||
binary_data_dir: data/xxx/binary
|
||||
binarization_args:
|
||||
num_workers: 0
|
||||
pe: parselmouth
|
||||
pe_ckpt: 'checkpoints/rmvpe/model.pt'
|
||||
hnsep: vr
|
||||
hnsep_ckpt: 'checkpoints/vr/model.pt'
|
||||
|
||||
use_lang_id: false
|
||||
num_lang: 1
|
||||
use_spk_id: false
|
||||
num_spk: 1
|
||||
# NOTICE: before enabling variance modules, please read the docs at
|
||||
# https://github.com/openvpi/DiffSinger/tree/main/docs/BestPractices.md#mutual-influence-between-variance-modules
|
||||
predict_dur: true
|
||||
predict_pitch: false
|
||||
# NOTICE: before enabling variance predictions, please read the docs at
|
||||
# https://github.com/openvpi/DiffSinger/tree/main/docs/BestPractices.md#choosing-variance-parameters
|
||||
predict_energy: false
|
||||
predict_breathiness: true
|
||||
predict_voicing: true
|
||||
predict_tension: false
|
||||
|
||||
energy_db_min: -96.0
|
||||
energy_db_max: -12.0
|
||||
|
||||
breathiness_db_min: -96.0
|
||||
breathiness_db_max: -20.0
|
||||
|
||||
voicing_db_min: -96.0
|
||||
voicing_db_max: -12.0
|
||||
|
||||
tension_logit_min: -10.0
|
||||
tension_logit_max: 10.0
|
||||
|
||||
enc_ffn_kernel_size: 3
|
||||
use_rope: true
|
||||
rope_interleaved: false
|
||||
use_stretch_embed: false
|
||||
use_variance_scaling: true
|
||||
hidden_size: 384
|
||||
dur_prediction_args:
|
||||
arch: resnet
|
||||
hidden_size: 256
|
||||
dropout: 0.1
|
||||
num_layers: 5
|
||||
kernel_size: 3
|
||||
log_offset: 1.0
|
||||
loss_type: mse
|
||||
lambda_pdur_loss: 0.3
|
||||
lambda_wdur_loss: 1.0
|
||||
lambda_sdur_loss: 3.0
|
||||
|
||||
use_melody_encoder: true
|
||||
melody_encoder_args:
|
||||
hidden_size: 128
|
||||
enc_layers: 4
|
||||
use_glide_embed: false
|
||||
glide_types: [up, down]
|
||||
glide_embed_scale: 11.313708498984760 # sqrt(128)
|
||||
|
||||
diffusion_type: reflow
|
||||
|
||||
pitch_prediction_args:
|
||||
pitd_norm_min: -8.0
|
||||
pitd_norm_max: 8.0
|
||||
pitd_clip_min: -12.0
|
||||
pitd_clip_max: 12.0
|
||||
repeat_bins: 64
|
||||
backbone_type: 'lynxnet2'
|
||||
backbone_args:
|
||||
num_layers: 6
|
||||
num_channels: 512
|
||||
dropout_rate: 0.0
|
||||
use_conditioner_cache: true
|
||||
glu_type: 'atanglu'
|
||||
|
||||
variances_prediction_args:
|
||||
total_repeat_bins: 72
|
||||
backbone_type: 'lynxnet2'
|
||||
backbone_args:
|
||||
num_layers: 6
|
||||
num_channels: 384
|
||||
dropout_rate: 0.0
|
||||
use_conditioner_cache: true
|
||||
glu_type: 'atanglu'
|
||||
|
||||
lambda_dur_loss: 1.0
|
||||
lambda_pitch_loss: 1.0
|
||||
lambda_var_loss: 1.0
|
||||
|
||||
optimizer_args:
|
||||
optimizer_cls: torch.optim.AdamW
|
||||
lr: 0.0006
|
||||
lr_scheduler_args:
|
||||
scheduler_cls: torch.optim.lr_scheduler.StepLR
|
||||
step_size: 5000
|
||||
gamma: 0.75
|
||||
max_batch_frames: 80000
|
||||
max_batch_size: 48
|
||||
max_updates: 60000
|
||||
|
||||
num_valid_plots: 10
|
||||
val_check_interval: 4000
|
||||
num_ckpt_keep: 8
|
||||
permanent_ckpt_start: 30000
|
||||
permanent_ckpt_interval: 10000
|
||||
pl_trainer_devices: 'auto'
|
||||
pl_trainer_precision: '16-mixed'
|
||||
@@ -0,0 +1,142 @@
|
||||
base_config:
|
||||
- configs/variance.yaml
|
||||
|
||||
dictionaries:
|
||||
zh: dictionaries/opencpop-extension.txt
|
||||
extra_phonemes: []
|
||||
merged_phoneme_groups: []
|
||||
|
||||
datasets:
|
||||
- raw_data_dir: data/xxx1/raw
|
||||
speaker: speaker1
|
||||
spk_id: 0
|
||||
language: zh
|
||||
test_prefixes:
|
||||
- wav1
|
||||
- wav2
|
||||
- wav3
|
||||
- wav4
|
||||
- wav5
|
||||
- raw_data_dir: data/xxx2/raw
|
||||
speaker: speaker2
|
||||
spk_id: 1
|
||||
language: zh
|
||||
test_prefixes:
|
||||
- wav1
|
||||
- wav2
|
||||
- wav3
|
||||
- wav4
|
||||
- wav5
|
||||
|
||||
binary_data_dir: data/xxx/binary
|
||||
binarization_args:
|
||||
num_workers: 0
|
||||
pe: parselmouth
|
||||
pe_ckpt: 'checkpoints/rmvpe/model.pt'
|
||||
hnsep: vr
|
||||
hnsep_ckpt: 'checkpoints/vr/model.pt'
|
||||
|
||||
use_lang_id: false
|
||||
num_lang: 1
|
||||
use_spk_id: false
|
||||
num_spk: 1
|
||||
# NOTICE: before enabling variance modules, please read the docs at
|
||||
# https://github.com/openvpi/DiffSinger/tree/main/docs/BestPractices.md#mutual-influence-between-variance-modules
|
||||
predict_dur: false
|
||||
predict_pitch: false
|
||||
# NOTICE: before enabling variance predictions, please read the docs at
|
||||
# https://github.com/openvpi/DiffSinger/tree/main/docs/BestPractices.md#choosing-variance-parameters
|
||||
predict_energy: false
|
||||
predict_breathiness: false
|
||||
predict_voicing: false
|
||||
predict_tension: false
|
||||
|
||||
energy_db_min: -96.0
|
||||
energy_db_max: -12.0
|
||||
|
||||
breathiness_db_min: -96.0
|
||||
breathiness_db_max: -20.0
|
||||
|
||||
voicing_db_min: -96.0
|
||||
voicing_db_max: -12.0
|
||||
|
||||
tension_logit_min: -10.0
|
||||
tension_logit_max: 10.0
|
||||
|
||||
enc_ffn_kernel_size: 3
|
||||
use_rope: true
|
||||
rope_interleaved: false
|
||||
use_stretch_embed: false
|
||||
use_variance_scaling: true
|
||||
hidden_size: 384
|
||||
dur_prediction_args:
|
||||
arch: resnet
|
||||
hidden_size: 256
|
||||
dropout: 0.1
|
||||
num_layers: 5
|
||||
kernel_size: 3
|
||||
log_offset: 1.0
|
||||
loss_type: mse
|
||||
lambda_pdur_loss: 0.3
|
||||
lambda_wdur_loss: 1.0
|
||||
lambda_sdur_loss: 3.0
|
||||
|
||||
use_melody_encoder: true
|
||||
melody_encoder_args:
|
||||
hidden_size: 128
|
||||
enc_layers: 4
|
||||
use_glide_embed: false
|
||||
glide_types: [up, down]
|
||||
glide_embed_scale: 11.313708498984760 # sqrt(128)
|
||||
|
||||
diffusion_type: reflow
|
||||
|
||||
pitch_prediction_args:
|
||||
pitd_norm_min: -8.0
|
||||
pitd_norm_max: 8.0
|
||||
pitd_clip_min: -12.0
|
||||
pitd_clip_max: 12.0
|
||||
repeat_bins: 64
|
||||
backbone_type: 'lynxnet2'
|
||||
backbone_args:
|
||||
num_layers: 6
|
||||
num_channels: 512
|
||||
dropout_rate: 0.0
|
||||
use_conditioner_cache: true
|
||||
glu_type: 'atanglu'
|
||||
|
||||
variances_prediction_args:
|
||||
total_repeat_bins: 72
|
||||
backbone_type: 'lynxnet2'
|
||||
backbone_args:
|
||||
num_layers: 6
|
||||
num_channels: 384
|
||||
dropout_rate: 0.0
|
||||
use_conditioner_cache: true
|
||||
glu_type: 'atanglu'
|
||||
|
||||
lambda_dur_loss: 1.0
|
||||
lambda_pitch_loss: 1.0
|
||||
lambda_var_loss: 1.0
|
||||
|
||||
optimizer_args:
|
||||
optimizer_cls: modules.optimizer.muon.Muon_AdamW
|
||||
lr: 0.0006
|
||||
muon_args:
|
||||
weight_decay: 0.1
|
||||
adamw_args:
|
||||
weight_decay: 0.0
|
||||
lr_scheduler_args:
|
||||
step_size: 5000
|
||||
gamma: 0.8
|
||||
max_batch_frames: 80000
|
||||
max_batch_size: 48
|
||||
max_updates: 80000
|
||||
|
||||
num_valid_plots: 10
|
||||
val_check_interval: 2000
|
||||
num_ckpt_keep: 8
|
||||
permanent_ckpt_start: 30000
|
||||
permanent_ckpt_interval: 10000
|
||||
pl_trainer_devices: 'auto'
|
||||
pl_trainer_precision: '16-mixed'
|
||||
@@ -0,0 +1,152 @@
|
||||
base_config:
|
||||
- configs/base.yaml
|
||||
|
||||
task_cls: training.variance_task.VarianceTask
|
||||
|
||||
dictionaries: {}
|
||||
extra_phonemes: []
|
||||
merged_phoneme_groups: []
|
||||
datasets: []
|
||||
|
||||
audio_sample_rate: 44100
|
||||
hop_size: 512 # Hop size.
|
||||
fft_size: 2048 # FFT size.
|
||||
win_size: 2048 # FFT size.
|
||||
midi_smooth_width: 0.06 # in seconds
|
||||
|
||||
binarization_args:
|
||||
shuffle: true
|
||||
num_workers: 0
|
||||
prefer_ds: false
|
||||
|
||||
binary_data_dir: 'data/opencpop_variance/binary'
|
||||
binarizer_cls: preprocessing.variance_binarizer.VarianceBinarizer
|
||||
|
||||
use_lang_id: false
|
||||
num_lang: 1
|
||||
use_spk_id: false
|
||||
num_spk: 1
|
||||
|
||||
predict_dur: true
|
||||
predict_pitch: true
|
||||
predict_energy: false
|
||||
predict_breathiness: false
|
||||
predict_voicing: false
|
||||
predict_tension: false
|
||||
|
||||
enc_ffn_kernel_size: 3
|
||||
use_rope: true
|
||||
rope_interleaved: false
|
||||
use_stretch_embed: false
|
||||
use_variance_scaling: true
|
||||
rel_pos: true
|
||||
hidden_size: 384
|
||||
|
||||
dur_prediction_args:
|
||||
arch: resnet
|
||||
hidden_size: 256
|
||||
dropout: 0.1
|
||||
num_layers: 5
|
||||
kernel_size: 3
|
||||
log_offset: 1.0
|
||||
loss_type: mse
|
||||
lambda_pdur_loss: 0.3
|
||||
lambda_wdur_loss: 1.0
|
||||
lambda_sdur_loss: 3.0
|
||||
|
||||
use_melody_encoder: true
|
||||
melody_encoder_args:
|
||||
hidden_size: 128
|
||||
enc_layers: 4
|
||||
use_glide_embed: false
|
||||
glide_types: [up, down]
|
||||
glide_embed_scale: 11.313708498984760 # sqrt(128)
|
||||
|
||||
pitch_prediction_args:
|
||||
pitd_norm_min: -8.0
|
||||
pitd_norm_max: 8.0
|
||||
pitd_clip_min: -12.0
|
||||
pitd_clip_max: 12.0
|
||||
repeat_bins: 64
|
||||
backbone_type: 'lynxnet2'
|
||||
backbone_args:
|
||||
num_layers: 6
|
||||
num_channels: 512
|
||||
dropout_rate: 0.0
|
||||
use_conditioner_cache: true
|
||||
glu_type: 'atanglu'
|
||||
|
||||
energy_db_min: -96.0
|
||||
energy_db_max: -12.0
|
||||
energy_smooth_width: 0.06
|
||||
|
||||
breathiness_db_min: -96.0
|
||||
breathiness_db_max: -20.0
|
||||
breathiness_smooth_width: 0.06
|
||||
voicing_db_min: -96.0
|
||||
voicing_db_max: -12.0
|
||||
voicing_smooth_width: 0.06
|
||||
|
||||
tension_logit_min: -10.0
|
||||
tension_logit_max: 10.0
|
||||
tension_smooth_width: 0.06
|
||||
|
||||
variances_prediction_args:
|
||||
total_repeat_bins: 72
|
||||
backbone_type: 'lynxnet2'
|
||||
backbone_args:
|
||||
num_layers: 6
|
||||
num_channels: 384
|
||||
dropout_rate: 0.0
|
||||
use_conditioner_cache: true
|
||||
glu_type: 'atanglu'
|
||||
|
||||
lambda_dur_loss: 1.0
|
||||
lambda_pitch_loss: 1.0
|
||||
lambda_var_loss: 1.0
|
||||
|
||||
diffusion_type: reflow # ddpm
|
||||
time_scale_factor: 1000
|
||||
schedule_type: 'linear'
|
||||
K_step: 1000
|
||||
timesteps: 1000
|
||||
max_beta: 0.02
|
||||
main_loss_type: l2
|
||||
main_loss_log_norm: true
|
||||
sampling_algorithm: euler
|
||||
sampling_steps: 20
|
||||
diff_accelerator: ddim
|
||||
diff_speedup: 10
|
||||
|
||||
# train and eval
|
||||
num_sanity_val_steps: 1
|
||||
optimizer_args:
|
||||
optimizer_cls: modules.optimizer.muon.Muon_AdamW
|
||||
lr: 0.0006
|
||||
muon_args:
|
||||
weight_decay: 0.1
|
||||
adamw_args:
|
||||
weight_decay: 0.0
|
||||
lr_scheduler_args:
|
||||
step_size: 5000
|
||||
gamma: 0.8
|
||||
max_batch_frames: 80000
|
||||
max_batch_size: 48
|
||||
dataset_size_key: 'lengths'
|
||||
val_check_interval: 4000
|
||||
num_valid_plots: 10
|
||||
max_updates: 80000
|
||||
num_ckpt_keep: 5
|
||||
permanent_ckpt_start: 30000
|
||||
permanent_ckpt_interval: 10000
|
||||
|
||||
finetune_enabled: false
|
||||
finetune_ckpt_path: null
|
||||
finetune_ignored_params:
|
||||
- model.spk_embed
|
||||
- model.fs2.txt_embed
|
||||
- model.fs2.encoder.embed_tokens
|
||||
finetune_strict_shapes: true
|
||||
|
||||
freezing_enabled: false
|
||||
frozen_params: []
|
||||
Reference in New Issue
Block a user