WaveGRU-Text-To-Speech / tacotron.toml
NTT123
a slow but working model
df1ad02
raw
history blame
No virus
738 Bytes
[tacotron]
# training
BATCH_SIZE = 64
LR=1024e-6 # learning rate
MODEL_PREFIX = "mono_tts_cbhg_small"
LOG_DIR = "./logs"
CKPT_DIR = "./ckpts"
USE_MP = false # use mixed-precision training
# data
TF_DATA_DIR = "./tf_data" # tensorflow data directory
TF_GTA_DATA_DIR = "./tf_gta_data" # tf gta data directory
SAMPLE_RATE = 24000 # convert to this sample rate if needed
MEL_DIM = 80 # the dimension of melspectrogram features
MEL_MIN = 1e-5
PAD = "_" # padding character
PAD_TOKEN = 0
TEST_DATA_SIZE = 1024
# model
RR = 2 # reduction factor
MAX_RR=2
ATTN_BIAS = 0.0 # control how slow the attention moves forward
SIGMOID_NOISE = 2.0
PRENET_DIM = 128
TEXT_DIM = 256
RNN_DIM = 512
ATTN_RNN_DIM = 256
ATTN_HIDDEN_DIM = 128
POSTNET_DIM = 512