File size: 2,957 Bytes
45ee559
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
import os

from trainer import Trainer, TrainerArgs

from TTS.config.shared_configs import BaseAudioConfig
from TTS.tts.configs.shared_configs import BaseDatasetConfig, CapacitronVAEConfig
from TTS.tts.configs.tacotron_config import TacotronConfig
from TTS.tts.datasets import load_tts_samples
from TTS.tts.models.tacotron import Tacotron
from TTS.tts.utils.text.tokenizer import TTSTokenizer
from TTS.utils.audio import AudioProcessor

output_path = os.path.dirname(os.path.abspath(__file__))

data_path = "/srv/data/"

# Using LJSpeech like dataset processing for the blizzard dataset
dataset_config = BaseDatasetConfig(formatter="ljspeech", meta_file_train="metadata.csv", path=data_path)

audio_config = BaseAudioConfig(
    sample_rate=24000,
    do_trim_silence=True,
    trim_db=60.0,
    signal_norm=True,
    mel_fmin=80.0,
    mel_fmax=12000,
    spec_gain=20.0,
    log_func="np.log10",
    ref_level_db=20,
    preemphasis=0.0,
    min_level_db=-100,
)

# Using the standard Capacitron config
capacitron_config = CapacitronVAEConfig(capacitron_VAE_loss_alpha=1.0)

config = TacotronConfig(
    run_name="Blizzard-Capacitron-T1",
    audio=audio_config,
    capacitron_vae=capacitron_config,
    use_capacitron_vae=True,
    batch_size=128,  # Tune this to your gpu
    max_audio_len=6 * 24000,  # Tune this to your gpu
    min_audio_len=0.5 * 24000,
    eval_batch_size=16,
    num_loader_workers=12,
    num_eval_loader_workers=8,
    precompute_num_workers=24,
    run_eval=True,
    test_delay_epochs=5,
    r=2,
    optimizer="CapacitronOptimizer",
    optimizer_params={"RAdam": {"betas": [0.9, 0.998], "weight_decay": 1e-6}, "SGD": {"lr": 1e-5, "momentum": 0.9}},
    attention_type="graves",
    attention_heads=5,
    epochs=1000,
    text_cleaner="phoneme_cleaners",
    use_phonemes=True,
    phoneme_language="en-us",
    phonemizer="espeak",
    phoneme_cache_path=os.path.join(data_path, "phoneme_cache"),
    stopnet_pos_weight=15,
    print_step=50,
    print_eval=True,
    mixed_precision=False,
    output_path=output_path,
    datasets=[dataset_config],
    lr=1e-3,
    lr_scheduler="StepwiseGradualLR",
    lr_scheduler_params={"gradual_learning_rates": [[0, 1e-3], [2e4, 5e-4], [4e4, 3e-4], [6e4, 1e-4], [8e4, 5e-5]]},
    scheduler_after_epoch=False,  # scheduler doesn't work without this flag
    loss_masking=False,
    decoder_loss_alpha=1.0,
    postnet_loss_alpha=1.0,
    postnet_diff_spec_alpha=1.0,
    decoder_diff_spec_alpha=1.0,
    decoder_ssim_alpha=1.0,
    postnet_ssim_alpha=1.0,
)

ap = AudioProcessor(**config.audio.to_dict())

tokenizer, config = TTSTokenizer.init_from_config(config)

train_samples, eval_samples = load_tts_samples(dataset_config, eval_split=True)

model = Tacotron(config, ap, tokenizer, speaker_manager=None)

trainer = Trainer(
    TrainerArgs(),
    config,
    output_path,
    model=model,
    train_samples=train_samples,
    eval_samples=eval_samples,
)

# πŸš€
trainer.fit()