lightspeech-mfa-sw-v1 / config.yml
w11wo's picture
Added Model
919b53f
allow_cache: true
batch_size: 8
config: ./TensorFlowTTS/examples/lightspeech/conf/lightspeech_swahiliipa.yaml
dataset_config: TensorFlowTTS/preprocess/swahiliipa_preprocess.yaml
dataset_mapping: dump/swahiliipa_mapper.json
dataset_stats: dump/stats.npy
delay_f0_energy_steps: 3
dev_dir: ./dump/valid/
energy_stat: ./dump/stats_energy.npy
eval_batch_size: 16
eval_interval_steps: 5000
f0_stat: ./dump/stats_f0.npy
format: npy
gradient_accumulation_steps: 2
hop_size: 512
is_shuffle: true
lightspeech_params:
attention_probs_dropout_prob: 0.1
dataset: swahiliipa
decoder_attention_head_size: 16
decoder_hidden_act: mish
decoder_hidden_size: 256
decoder_intermediate_kernel_size:
- 17
- 21
- 9
- 13
decoder_intermediate_size: 1024
decoder_num_attention_heads: 2
decoder_num_hidden_layers: 3
encoder_attention_head_size: 16
encoder_hidden_act: mish
encoder_hidden_size: 256
encoder_intermediate_kernel_size:
- 5
- 25
- 13
- 9
encoder_intermediate_size: 1024
encoder_num_attention_heads: 2
encoder_num_hidden_layers: 3
hidden_dropout_prob: 0.2
initializer_range: 0.02
max_position_embeddings: 2048
n_speakers: 1
num_mels: 80
output_attentions: false
output_hidden_states: false
variant_prediction_num_conv_layers: 2
variant_predictor_dropout_rate: 0.5
variant_predictor_filter: 256
variant_predictor_kernel_size: 3
log_interval_steps: 200
mel_length_threshold: 32
mixed_precision: true
model_type: lightspeech
num_save_intermediate_results: 1
optimizer_params:
decay_steps: 150000
end_learning_rate: 5.0e-05
initial_learning_rate: 0.001
warmup_proportion: 0.02
weight_decay: 0.001
outdir: ./lightspeech-openbible
pretrained: ''
remove_short_samples: true
resume: ''
save_interval_steps: 5000
train_dir: ./dump/train/
train_max_steps: 200000
use_norm: true
var_train_expr: null
verbose: 1
version: '0.0'