Different sampling rates
Opened this issue · 4 comments
anhnv125 commented
Hi,
I would like to train the model to upsample from 8 kHz to 16 kHz. What should I change?
Thank you!
EmreOzkose commented
I am also curious about that.
EmreOzkose commented
@anhnv125 Have you tried to train a model?
anhnv125 commented
@EmreOzkose Yes, I fixed some bugs and changed the implementation into Lightning. You can find the results in my paper and repo. Here is my config file
audio_num_mel_bins: 80
audio_sample_rate: 16000
#base_config:
#- configs/tts/vctk/resample.yaml
binarization_args:
resample_ratio: 0.5
shuffle: false
with_align: false
with_f0: true
with_f0cwt: true
with_resample: true
with_spk_embed: false
with_txt: true
with_wav: true
#binarizer_cls: data_gen.tts.base_binarizer.BaseBinarizer
binary_data_dir: data/binary/vctk_wav
check_val_every_n_epoch: 10
clip_grad_norm: 1
debug: false
dec_ffn_kernel_size: 9
dec_layers: 4
dict_dir: ''
disc_start_steps: 40000
discriminator_grad_norm: 1
discriminator_optimizer_params:
eps: 1.0e-06
lr: 5.0e-05
weight_decay: 0.0
discriminator_params:
bias: true
conv_channels: 64
in_channels: 1
kernel_size: 3
layers: 10
nonlinear_activation: LeakyReLU
nonlinear_activation_params:
negative_slope: 0.2
out_channels: 1
use_weight_norm: true
discriminator_scheduler_params:
gamma: 0.5
step_size: 200000
dropout: 0.1
ds_dworkers: 8
enc_ffn_kernel_size: 9
enc_layers: 4
endless_ds: true
ffn_act: gelu
ffn_padding: SAME
fft_size: 1024
fmax: 7600
fmin: 80
format: hdf5
gen_dir_name: ''
generator_grad_norm: 10
generator_optimizer_params:
eps: 1.0e-06
lr: 0.0001
weight_decay: 0.0
generator_params:
aux_channels: 80
aux_context_window: 2
dropout: 0.0
gate_channels: 128
in_channels: 1
kernel_size: 3
layers: 30
out_channels: 1
residual_channels: 64
skip_channels: 64
stacks: 3
upsample_net: ConvInUpsampleNetwork
upsample_params:
upsample_scales:
- 4
- 4
- 4
- 4
use_pitch_embed: false
use_weight_norm: true
generator_scheduler_params:
gamma: 0.5
step_size: 200000
griffin_lim_iters: 60
hidden_size: 384
hop_size: 256
infer: false
is_freqwg: false
islb2lr: false
lambda_adv: 4.0
load_ckpt: ''
log_interval: 100
loud_norm: false
lr: 1e-5
max_epochs: 1000
max_eval_sentences: 5
max_eval_tokens: 60000
max_frames: 1550
max_input_tokens: 400
max_samples: 12800
max_sentences: 36
max_tokens: 30000
max_updates: 1000000
mel_vmax: 1.5
mel_vmin: -6
min_level_db: -100
model_class: WaveGlowMelHF
n_fft: 32
num_ckpt_keep: 3
num_heads: 2
num_mels: 80
num_sanity_val_steps: 5
num_spk: 400
num_test_samples: 5
num_valid_plots: 5
optimizer_adam_beta1: 0.9
optimizer_adam_beta2: 0.98
out_wav_norm: false
pre_align_args:
allow_no_txt: false
denoise: false
forced_align: mfa
txt_processor: en
use_sox: false
use_tone: true
pre_align_cls: ''
prenet_dropout: 0.5
prenet_hidden_size: 256
print_nan_grads: false
processed_data_dir: data/processed/vctk
profile_infer: false
raw_data_dir: data
ref_level_db: 20
reset_phone_dict: true
sampling_rate: 16000
save_best: true
save_ckpt: true
save_codes:
- configs
- modules
- tasks
- utils
- usr
save_f0: false
save_gt: true
seed: 1234
sort_by_len: true
stft_loss_params:
fft_sizes:
- 1024
- 2048
- 512
hop_sizes:
- 120
- 240
- 50
win_lengths:
- 600
- 1200
- 240
window: hann_window
stop_token_weight: 5.0
task_cls: tasks.super_resolution.waveglow_hf.WaveGlowHFTask
test_ids:
- 12
- 23
- 34
- 45
- 56
- 67
- 70
- 74
- 87
- 99
test_input_dir: ''
test_num: 100
test_set_name: test
train_set_name: train
use_mel_loss: false
val_check_interval: 200
valid_set_name: valid
vocoder: pwg
vocoder_ckpt: ''
warmup_updates: 8000
waveglow_config:
WN_config:
kernel_size: 3
n_channels: 256
n_layers: 8
embed_dim: 400
embed_num: 256
mu: 256
n_early_every: 4
n_early_size: 2
n_flows: 12
n_group: 8
weight_decay: 0
win_length: null
win_size: 1024
window: hann
work_dir: /tf/WSRGlow/checkpoints/wsrglow
EmreOzkose commented
Thank you so much