Create config.json
Browse files- config.json +64 -0
config.json
ADDED
|
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"allow_cache": true,
|
| 3 |
+
"batch_size": 16,
|
| 4 |
+
"config": "/content/TensorFlowTTS/examples/fastspeech2/conf/fastspeech2.v1.yaml",
|
| 5 |
+
"dev_dir": "/content/dump_synpaflex/valid/",
|
| 6 |
+
"energy_stat": "/content/dump_synpaflex/stats_energy.npy",
|
| 7 |
+
"eval_interval_steps": 500,
|
| 8 |
+
"f0_stat": "/content/dump_synpaflex/stats_f0.npy",
|
| 9 |
+
"fastspeech2_params": {
|
| 10 |
+
"attention_probs_dropout_prob": 0.1,
|
| 11 |
+
"decoder_attention_head_size": 192,
|
| 12 |
+
"decoder_hidden_act": "mish",
|
| 13 |
+
"decoder_hidden_size": 384,
|
| 14 |
+
"decoder_intermediate_kernel_size": 3,
|
| 15 |
+
"decoder_intermediate_size": 1024,
|
| 16 |
+
"decoder_num_attention_heads": 2,
|
| 17 |
+
"decoder_num_hidden_layers": 4,
|
| 18 |
+
"encoder_attention_head_size": 192,
|
| 19 |
+
"encoder_hidden_act": "mish",
|
| 20 |
+
"encoder_hidden_size": 384,
|
| 21 |
+
"encoder_intermediate_kernel_size": 3,
|
| 22 |
+
"encoder_intermediate_size": 1024,
|
| 23 |
+
"encoder_num_attention_heads": 2,
|
| 24 |
+
"encoder_num_hidden_layers": 4,
|
| 25 |
+
"hidden_dropout_prob": 0.2,
|
| 26 |
+
"initializer_range": 0.02,
|
| 27 |
+
"max_position_embeddings": 2048,
|
| 28 |
+
"n_speakers": 1,
|
| 29 |
+
"num_mels": 80,
|
| 30 |
+
"output_attentions": false,
|
| 31 |
+
"output_hidden_states": false,
|
| 32 |
+
"variant_prediction_num_conv_layers": 2,
|
| 33 |
+
"variant_predictor_dropout_rate": 0.5,
|
| 34 |
+
"variant_predictor_filter": 256,
|
| 35 |
+
"variant_predictor_kernel_size": 3
|
| 36 |
+
},
|
| 37 |
+
"format": "npy",
|
| 38 |
+
"gradient_accumulation_steps": 1,
|
| 39 |
+
"hop_size": 256,
|
| 40 |
+
"is_shuffle": true,
|
| 41 |
+
"log_interval_steps": 200,
|
| 42 |
+
"mel_length_threshold": 32,
|
| 43 |
+
"mixed_precision": true,
|
| 44 |
+
"model_type": "fastspeech2",
|
| 45 |
+
"num_save_intermediate_results": 1,
|
| 46 |
+
"optimizer_params": {
|
| 47 |
+
"decay_steps": 150000,
|
| 48 |
+
"end_learning_rate": 5e-05,
|
| 49 |
+
"initial_learning_rate": 0.001,
|
| 50 |
+
"warmup_proportion": 0.02,
|
| 51 |
+
"weight_decay": 0.001
|
| 52 |
+
},
|
| 53 |
+
"outdir": "/content/drive/MyDrive/exp/train.fastspeech2.v1/",
|
| 54 |
+
"pretrained": "",
|
| 55 |
+
"remove_short_samples": true,
|
| 56 |
+
"resume": "/content/drive/MyDrive/exp/train.fastspeech2.v1/checkpoints/ckpt-1.index",
|
| 57 |
+
"save_interval_steps": 5000,
|
| 58 |
+
"train_dir": "/content/dump_synpaflex/train/",
|
| 59 |
+
"train_max_steps": 200000,
|
| 60 |
+
"use_norm": true,
|
| 61 |
+
"var_train_expr": null,
|
| 62 |
+
"verbose": 1,
|
| 63 |
+
"version": "0.0"
|
| 64 |
+
}
|