74 lines
1.8 KiB
YAML
74 lines
1.8 KiB
YAML
|
#### general settings
|
||
|
name: test_tacotron2_lj
|
||
|
use_tb_logger: true
|
||
|
gpu_ids: [0]
|
||
|
start_step: -1
|
||
|
fp16: false
|
||
|
checkpointing_enabled: true
|
||
|
wandb: false
|
||
|
|
||
|
datasets:
|
||
|
train:
|
||
|
name: lj
|
||
|
n_workers: 0
|
||
|
batch_size: 1
|
||
|
mode: nv_tacotron
|
||
|
path: E:\4k6k\datasets\audio\LJSpeech-1.1\ljs_audio_text_train_filelist.txt
|
||
|
|
||
|
networks:
|
||
|
mel_gen:
|
||
|
type: generator
|
||
|
which_model_G: nv_tacotron2
|
||
|
args:
|
||
|
encoder_kernel_size: 5
|
||
|
encoder_n_convolutions: 3
|
||
|
encoder_embedding_dim: 512
|
||
|
decoder_rnn_dim: 1024
|
||
|
prenet_dim: 256
|
||
|
max_decoder_steps: 1000
|
||
|
attention_rnn_dim: 1024
|
||
|
attention_dim: 128
|
||
|
attention_location_n_filters: 32
|
||
|
attention_location_kernel_size: 31
|
||
|
postnet_embedding_dim: 512
|
||
|
postnet_kernel_size: 5
|
||
|
postnet_n_convolutions: 5
|
||
|
waveglow:
|
||
|
type: generator
|
||
|
which_model_G: nv_waveglow
|
||
|
args:
|
||
|
n_mel_channels: 80
|
||
|
n_flows: 12
|
||
|
n_group: 8
|
||
|
n_early_every: 4
|
||
|
n_early_size: 2
|
||
|
WN_config:
|
||
|
n_layers: 8
|
||
|
n_channels: 256
|
||
|
kernel_size: 3
|
||
|
|
||
|
#### path
|
||
|
path:
|
||
|
pretrain_model_mel_gen: ../experiments/train_tacotron2_lj/models/22000_mel_gen_ema.pth
|
||
|
pretrain_model_waveglow: ../experiments/waveglow_256channels_universal_v5.pth
|
||
|
strict_load: true
|
||
|
#resume_state: ../experiments/train_imgset_unet_diffusion/training_state/54000.state
|
||
|
|
||
|
steps:
|
||
|
generator:
|
||
|
training: mel_gen
|
||
|
injectors:
|
||
|
mel:
|
||
|
type: generator
|
||
|
generator: mel_gen
|
||
|
in: [padded_text, input_lengths, padded_mel, output_lengths]
|
||
|
out: [mel_outputs, mel_outputs_postnet, gate_outputs, alignments]
|
||
|
wave:
|
||
|
type: generator
|
||
|
generator: waveglow
|
||
|
method: infer
|
||
|
in: mel_outputs
|
||
|
out: waveform
|
||
|
|
||
|
eval:
|
||
|
output_state: waveform
|