mirror of https://github.com/coqui-ai/TTS.git
127 lines
3.2 KiB
JSON
127 lines
3.2 KiB
JSON
{
|
|
"attention_heads": 4,
|
|
"attention_norm": "sigmoid",
|
|
"attention_type": "original",
|
|
"audio_config": {
|
|
"clip_norm": true,
|
|
"do_trim_silence": true,
|
|
"fft_size": 1024,
|
|
"frame_length_ms": null,
|
|
"frame_shift_ms": null,
|
|
"griffin_lim_iters": 60,
|
|
"hop_length": 256,
|
|
"max_norm": 4,
|
|
"mel_fmax": 7600,
|
|
"mel_fmin": 50,
|
|
"min_level_db": -100,
|
|
"num_mels": 80,
|
|
"power": 1.5,
|
|
"preemphasis": 0,
|
|
"ref_level_db": 20,
|
|
"sample_rate": 22050,
|
|
"signal_norm": true,
|
|
"spec_gain": 1,
|
|
"stats_path": "/home/erogol/Data/LJSpeech-1.1/scale_stats.npy",
|
|
"symmetric_norm": true,
|
|
"trim_db": 60,
|
|
"win_length": 1024
|
|
},
|
|
"bidirectional_decoder": false,
|
|
"compute_input_seq_cache": false,
|
|
"ddc_r": 7,
|
|
"decoder_diff_spec_alpha": 0.25,
|
|
"decoder_loss_alpha": 0.5,
|
|
"decoder_ssim_alpha": 0.5,
|
|
"double_decoder_consistency": true,
|
|
"enable_eos_bos_chars": false,
|
|
"forward_attn_mask": false,
|
|
"ga_alpha": 5,
|
|
"grad_clip": 1,
|
|
"gradual_training": [
|
|
[
|
|
0,
|
|
7,
|
|
64
|
|
],
|
|
[
|
|
1,
|
|
5,
|
|
64
|
|
],
|
|
[
|
|
50000,
|
|
3,
|
|
32
|
|
],
|
|
[
|
|
130000,
|
|
2,
|
|
32
|
|
],
|
|
[
|
|
290000,
|
|
1,
|
|
32
|
|
]
|
|
],
|
|
"location_attn": true,
|
|
"lr": 0.0001,
|
|
"memory_size": -1,
|
|
"noam_schedule": false,
|
|
"phoneme_cache_path": "/home/erogol/Models/phoneme_cache/",
|
|
"phoneme_language": "en-us",
|
|
"postnet_diff_spec_alpha": 0.25,
|
|
"postnet_loss_alpha": 0.25,
|
|
"postnet_ssim_alpha": 0.25,
|
|
"prenet_dropout": false,
|
|
"prenet_type": "original",
|
|
"r": 7,
|
|
"separate_stopnet": true,
|
|
"seq_len_norm": false,
|
|
"stopnet": true,
|
|
"stopnet_pos_weight": 15,
|
|
"test_sentences_file": null,
|
|
"text_cleaner": "phoneme_cleaners",
|
|
"training_config": {
|
|
"batch_group_size": 4,
|
|
"batch_size": 32,
|
|
"checkpoint": true,
|
|
"datasets": [
|
|
{
|
|
"meta_file_train": "metadata.csv",
|
|
"meta_file_val": null,
|
|
"name": "ljspeech",
|
|
"path": "/home/erogol/Data/LJSpeech-1.1/"
|
|
}
|
|
],
|
|
"epochs": 1000,
|
|
"eval_batch_size": 16,
|
|
"keep_after": 10000,
|
|
"keep_all_best": false,
|
|
"loss_masking": true,
|
|
"max_seq_len": 153,
|
|
"min_seq_len": 6,
|
|
"mixed_precision": true,
|
|
"model": "Tacotron2",
|
|
"num_loader_workers": 4,
|
|
"num_val_loader_workers": 4,
|
|
"output_path": "/home/erogol/Models/LJSpeech/",
|
|
"print_eval": false,
|
|
"print_step": 25,
|
|
"run_description": "tacotron2 with DDC and differential spectral loss.",
|
|
"run_eval": true,
|
|
"run_name": "ljspeech-ddc",
|
|
"save_step": 10000,
|
|
"tb_model_param_stats": false,
|
|
"tb_plot_step": 100,
|
|
"test_delay_epochs": 10,
|
|
"use_noise_augment": true
|
|
},
|
|
"transition_agent": false,
|
|
"use_forward_attn": false,
|
|
"use_phonemes": true,
|
|
"warmup_steps": 4000,
|
|
"wd": 0.000001,
|
|
"windowing": false
|
|
}
|