tcsinger.yaml

base_config: egs/egs_bases/tts/fs.yaml
task_cls: tasks.TCSinger.tcsinger.TCSingerTask
binary_data_dir: 'data/binary/tc'
processed_data_dir: 'data/processed/tc'
binarizer_cls: data_gen.tc_binarizer.TCBinarizer

vocoder: HifiGAN_NSF
vocoder_ckpt: checkpoints/hifigan
use_nsf: true

ds_workers: 8
max_input_tokens: 5625
max_tokens: 20000

val_check_interval: 5000
valid_infer_interval: 5000
max_updates: 310000

audio_sample_rate: 48000
hop_size: 256
win_size: 1024
fft_size: 1024
fmax: 24000
fmin: 20
max_frames: 3000
f0_max: 1000
f0_min: 50
num_spk: 150
use_spk_id: false
use_spk_embed: false

# style encoder related
vq: 'cvq' # or ema
hidden_size: 320
ffn_hidden_size: 1024
vq_input_dim: 80
vq_ph_codebook_dim: 512
vq_ph_channel: 64
vq_ph_beta: 0.25

# Text encoder type 
encoder_type: rel_fft

# f0 related
use_pitch_embed: true
f0_gen: gmdiff
param_: eps
f0_timesteps: 100
f0_K_step: 100
f0_infer_with_ref: false

# decoder related
de: 'sad'
gaussian_start: true
f0_dilation_cycle_length: 4  # *
predictor_layers: 5
diff_loss_type: l1
f0_max_beta: 0.06
f0_residual_layers: 10
f0_residual_channels: 192
optimizer_adam_beta1: 0.9
optimizer_adam_beta2: 0.98
weight_decay: 0
keep_bins: 80
diff_decoder_type: 'sad'
max_beta: 0.06
timesteps: 8
timescale: 1
schedule_type: 'vpsde'
dilation_cycle_length: 1
residual_layers: 20
residual_channels: 320
spec_min: [ ]
spec_max: [ ]

binarization_args:
  reset_phone_dict: true
  reset_word_dict: true
  shuffle: false
  trim_eos_bos: false
  trim_sil: false
  with_align: true
  with_f0: true
  with_f0cwt: false
  with_linear: false
  with_spk_embed: false
  with_spk_id: true
  with_txt: true
  with_wav: true
  with_word: false

num_sanity_val_steps: 0 # steps of validation at the beginning

# dataset split
valid_prefixes: [

]
test_prefixes: [

]

test_ids: []