38 lines
938 B
YAML
38 lines
938 B
YAML
# Path
|
|
dataset_path: "/workspace/fastspeech/LJSpeech-1.1"
|
|
tacotron2_path: "/workspace/fastspeech/tacotron2_statedict.pt"
|
|
waveglow_path: "/workspace/fastspeech/nvidia_waveglow256pyt_fp16"
|
|
mels_path: "/workspace/fastspeech/mels_ljspeech1.1"
|
|
aligns_path: "/workspace/fastspeech/aligns_ljspeech1.1"
|
|
log_path: "/workspace/fastspeech/logs"
|
|
checkpoint_path: "/workspace/fastspeech/checkpoints"
|
|
|
|
# Audio
|
|
sr: 22050
|
|
n_fft: 1024
|
|
win_len: 1024
|
|
hop_len: 256
|
|
num_mels: 80
|
|
mel_fmin: 0.0
|
|
mel_fmax: 8000.0
|
|
|
|
# Text
|
|
text_cleaners: ['english_cleaners']
|
|
|
|
# Model
|
|
d_model: 384
|
|
phoneme_side_n_layer: 6
|
|
phoneme_side_head: 2
|
|
phoneme_side_conv1d_filter_size: 1536
|
|
max_seq_len: 2048 # 23s
|
|
phoneme_side_output_size: 384
|
|
mel_side_n_layer: 6
|
|
mel_side_head: 2
|
|
mel_side_conv1d_filter_size: 1536
|
|
mel_side_output_size: 384
|
|
fft_conv1d_kernel: 3
|
|
fft_conv1d_padding: 1
|
|
duration_predictor_filter_size: 256
|
|
duration_predictor_kernel_size: 3
|
|
dropout: 0.1
|
|
fused_layernorm: False |