config/config_spect_c16_ssc.yaml

data:
  root_dir: '/u/wjkang/data/VCTK-Corpus/VCTK-Corpus' # root path of train data (either relative/absolute path is ok)
  wav_dir: 'wav16'
  spect_dir: 'spect'
  f0_norm_dir: 'f0_norm'
  avg_speaker_embs_file: '/u/wjkang/data/VCTK-Corpus/VCTK-Corpus/metadata/fast_resnet34_avg_embs.pkl'
  speaker_embs_gmm_file: '/u/wjkang/data/VCTK-Corpus/VCTK-Corpus/metadata/fast_resnet34_emb_gmms.pkl'
  seen_speakers_train_utts: '/u/wjkang/data/VCTK-Corpus/VCTK-Corpus/metadata/seen_speakers_train_utts.pkl'
  seen_speakers_test_utts: '/u/wjkang/data/VCTK-Corpus/VCTK-Corpus/metadata/seen_speakers_test_utts.pkl'
  f0_metadata_file: '/u/wjkang/data/VCTK-Corpus/VCTK-Corpus/metadata/speaker_f0_metadata.pkl'
#############################
train:
  num_workers: 8
  num_gpus: 1
  batch_size: 16
  optimizer: 'adam'
  seed: 1234
  adam:
    lr: 0.00005
    beta1: 0.5
    beta2: 0.9
  stft_lamb: 2.5
  use_wav2vec: False
  use_gmm_emb: True
  warp_lq: True
  use_ssc: True
#############################
audio:
  feat_dim: 80
  n_mel_channels: 80
  f0_norm_dim: 257
  spk_emb_dim: 512
  spk_quant_f0_dim: 64
  segment_length: 16384
  pad_short: 2000
  filter_length: 1024
  hop_length: 256
  win_length: 1024
  wav2vec_hop_length: None
  sampling_rate: 16000
  mel_fmin: 0.0
  mel_fmax: 8000.0
#############################
gen:
  noise_dim: 64
  channel_size: 16
  dilations: [1, 3, 9, 27]
  strides: [8, 8, 4]
  lReLU_slope: 0.2
  kpnet_conv_size: 3
#############################
ssc:
  se:
    spk_emb_dim: 512
    num_filters: [16, 32, 64, 128]
    layers: [3, 4, 6, 3]
    pretrained_weight_path: "./weights/resnet34sel_pretrained.pt"
  stft_annealing_step: 2000
  pos_ssc_lamb: 0.9
  neg_ssc_lamb: 0.0
  ssc_annealing_step: 2000
  num_ssc_samples: 8
  finetune_epochs: 3
#############################
mpd:
  periods: [2, 3, 5, 7, 11]
  kernel_size: 5
  stride: 3
  use_spectral_norm: False
  lReLU_slope: 0.2
#############################
mrd:
  resolutions: "[(5, 25), (10, 50), (2, 10)]" # (hop_length_ms, win_length_ms)
  use_spectral_norm: False
  lReLU_slope: 0.2
#############################
log:
  summary_interval: 10
  validation_interval: 1
  save_interval: 1
  num_audio: 5
  chkpt_dir: 'chkpt'
  log_dir: 'logs'
  ssc_validation_interval_steps: 400