params.py

from types import SimpleNamespace

################################
# Experiment Parameters        #
################################
# epochs=500,
# iters_per_checkpoint=1000,
# seed=1234,
# dynamic_loss_scaling=True,
# fp16_run=False,
# distributed_run=False,
# dist_backend="nccl",
# dist_url="tcp://localhost:54321",
# cudnn_enabled=True,
# cudnn_benchmark=False,
# ignore_layers=['embedding.weight'],

################################
# Data Parameters             #
################################
# load_mel_from_disk=False,
# training_files='filelists/ljs_audio_text_train_filelist.txt',
# validation_files='filelists/ljs_audio_text_val_filelist.txt',
# text_cleaners=['english_cleaners'],

tacotron_params = {
    ################################
    # Audio Parameters             #
    ################################
    'max_wav_value': 32768.0,  # for 16 bits files
    'sampling_rate': 22050,
    'filter_length': 1024,
    'hop_length': 256,
    'win_length': 1024,
    'n_mel_channels': 80,
    'mel_fmin': 0.0,
    'mel_fmax': 8000.0,

    ################################
    # Model Parameters             #
    ################################
    #n_symbols=len(symbols),
    #symbols_embedding_dim=512,

    # Encoder parameters
    'encoder_kernel': 5,
    'encoder_n_conv': 3,
    'encoder_embedding_dim': 512,

    # Decoder parameters
    'n_frames_per_step': 1,  # currently only 1 is supported
    'decoder_rnn_dim': 1024,
    'prenet_dim': 256,
    'max_decoder_steps': 1000,
    'gate_threshold': 0.5,
    'p_attention_dropout': 0.1,
    'p_decoder_dropout': 0.1,

    # Attention parameters
    'attn_rnn_dim': 1024,
    'attn_dim': 128,

    # Location Layer parameters
    'attn_loc_n_filters': 32,
    'attn_loc_kernel': 31,

    # Mel-post processing network parameters
    'postnet_embedding_dim': 512,
    'postnet_kernel_size': 5,
    'postnet_n_convolutions': 5
}

tacotron_params = SimpleNamespace(**tacotron_params)
################################
# Optimization Hyperparameters #
################################
# use_saved_learning_rate=False,
# learning_rate=1e-3,
# weight_decay=1e-6,
# grad_clip_thresh=1.0,
# batch_size=64,
# mask_padding=True  # set model's padded outputs to padded values