|
|
|
|
|
|
|
|
|
###########################################################
|
|
|
|
|
# FEATURE EXTRACTION SETTING #
|
|
|
|
|
###########################################################
|
|
|
|
|
fs: 24000 # Sampling rate.
|
|
|
|
|
n_fft: 2048 # FFT size (samples).
|
|
|
|
|
n_shift: 300 # Hop size (samples). 12.5ms
|
|
|
|
|
win_length: 1200 # Window length (samples). 50ms
|
|
|
|
|
# If set to null, it will be the same as fft_size.
|
|
|
|
|
window: "hann" # Window function.
|
|
|
|
|
n_mels: 80 # Number of mel basis.
|
|
|
|
|
fmin: 80 # Minimum freq in mel basis calculation. (Hz)
|
|
|
|
|
fmax: 7600 # Maximum frequency in mel basis calculation. (Hz)
|
|
|
|
|
mu_law: True # Recommended to suppress noise if using raw bitsexit()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
###########################################################
|
|
|
|
|
# MODEL SETTING #
|
|
|
|
|
###########################################################
|
|
|
|
|
model:
|
|
|
|
|
rnn_dims: 512 # Hidden dims of RNN Layers.
|
|
|
|
|
fc_dims: 512
|
|
|
|
|
bits: 9 # Bit depth of signal
|
|
|
|
|
aux_context_window: 2 # Context window size for auxiliary feature.
|
|
|
|
|
# If set to 2, previous 2 and future 2 frames will be considered.
|
|
|
|
|
aux_channels: 80 # Number of channels for auxiliary feature conv.
|
|
|
|
|
# Must be the same as num_mels.
|
|
|
|
|
upsample_scales: [4, 5, 3, 5] # Upsampling scales. Prodcut of these must be the same as hop size, same with pwgan here
|
|
|
|
|
compute_dims: 128 # Dims of Conv1D in MelResNet.
|
|
|
|
|
res_out_dims: 128 # Dims of output in MelResNet.
|
|
|
|
|
res_blocks: 10 # Number of residual blocks.
|
|
|
|
|
mode: RAW # either 'raw'(softmax on raw bits) or 'mold' (sample from mixture of logistics)
|
|
|
|
|
inference:
|
|
|
|
|
gen_batched: True # whether to genenate sample in batch mode
|
|
|
|
|
target: 12000 # target number of samples to be generated in each batch entry
|
|
|
|
|
overlap: 600 # number of samples for crossfading between batches
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
###########################################################
|
|
|
|
|
# DATA LOADER SETTING #
|
|
|
|
|
###########################################################
|
|
|
|
|
batch_size: 64 # Batch size.
|
|
|
|
|
batch_max_steps: 4500 # Length of each audio in batch. Make sure dividable by hop_size.
|
|
|
|
|
num_workers: 2 # Number of workers in DataLoader.
|
|
|
|
|
|
|
|
|
|
###########################################################
|
|
|
|
|
# OPTIMIZER SETTING #
|
|
|
|
|
###########################################################
|
|
|
|
|
grad_clip: 4.0
|
|
|
|
|
learning_rate: 1.0e-4
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
###########################################################
|
|
|
|
|
# INTERVAL SETTING #
|
|
|
|
|
###########################################################
|
|
|
|
|
|
|
|
|
|
train_max_steps: 400000 # Number of training steps.
|
|
|
|
|
save_interval_steps: 5000 # Interval steps to save checkpoint.
|
|
|
|
|
eval_interval_steps: 1000 # Interval steps to evaluate the network.
|
|
|
|
|
gen_eval_samples_interval_steps: 5000 # the iteration interval of generating valid samples
|
|
|
|
|
generate_num: 5 # number of samples to generate at each checkpoint
|
|
|
|
|
|
|
|
|
|
###########################################################
|
|
|
|
|
# OTHER SETTING #
|
|
|
|
|
###########################################################
|
|
|
|
|
num_snapshots: 10 # max number of snapshots to keep while training
|
|
|
|
|
seed: 42 # random seed for paddle, random, and np.random
|
