spellbrush/AliasingFreeNeuralAudioSynthesis

1{2    "base_config": "egs/afgen/exp_config_afcodec_base.json",3    "dataset": [4        "afgen_all_data",5    ],6    "exp_name": "afcodec",7    "inference": {8        "batch_size": 1,9    },10    "log_dir": "/home/vocod/experiments/afgen",11    "model": {12        "afcodec": {13            "codebook_dim": 8,14            "codebook_size": 1024,15            "decoder_dim": 512,16            "decoder_rates": [17                8,18                8,19                2,20                2,21                2,22            ],23            "encoder_dim": 32,24            "encoder_rates": [25                2,26                2,27                2,28                8,29                8,30            ],31            "n_codebooks": 9,32            "quantizer_dropout": 0.5,33        },34        "discriminators": [35            "msd",36            "mpd",37            "mssbstftd",38            "mssbcqtd",39        ],40        "generator": "afcodec",41        "mpd": {42            "discriminator_channel_mult_factor": 1,43            "mpd_reshapes": [44                2,45                3,46                5,47                7,48                11,49                17,50                23,51                37,52            ],53            "use_spectral_norm": false,54        },55        "mssbcqtd": {56            "bins_per_octaves": [57                24,58                36,59                48,60            ],61            "dilations": [62                1,63                2,64                4,65            ],66            "filters": 32,67            "filters_scale": 1,68            "hop_lengths": [69                1024,70                512,71                512,72            ],73            "in_channels": 1,74            "max_filters": 1024,75            "n_octaves": [76                10,77                10,78                10,79            ],80            "out_channels": 1,81        },82        "msstftd": {83            "filters": 32,84        },85    },86    "model_type": "AFGenCodec",87    "preprocess": {88        "align_mel_duration": false,89        "audio_dir": "audios",90        "bits": 8,91        "contentvec_dir": "contentvec",92        "cut_mel_frame": 128,93        "data_augment": false,94        "dur_dir": "durs",95        "duration_dir": "duration",96        "emo2id": "emo2id.json",97        "energy_dir": "energys",98        "energy_extract_mode": "from_mel",99        "energy_norm": false,100        "energy_remove_outlier": false,101        "extract_acoustic_token": false,102        "extract_amplitude_phase": false,103        "extract_audio": true,104        "extract_contentvec_feature": false,105        "extract_duration": false,106        "extract_energy": false,107        "extract_label": false,108        "extract_linear_spec": false,109        "extract_mcep": false,110        "extract_mel": false,111        "extract_mert_feature": false,112        "extract_one_hot": false,113        "extract_phone": false,114        "extract_pitch": true,115        "extract_uv": true,116        "extract_wenet_feature": false,117        "extract_whisper_feature": false,118        "f0_max": 1975.5,119        "f0_min": 32.7,120        "file_lst": "file.lst",121        "fmax": 22050,122        "fmin": 0,123        "hop_size": 512,124        "imaginary_dir": "imaginarys",125        "is_mu_law": false,126        "lab_dir": "labs",127        "label_dir": "labels",128        "lexicon_path": "./text/lexicon/librispeech-lexicon.txt",129        "linear_dir": "linears",130        "log_amplitude_dir": "log_amplitudes",131        "mcep_dir": "mcep",132        "mel_dir": "mels",133        "mel_extract_mode": "",134        "mel_min_max_norm": false,135        "min_level_db": -115,136        "n_fft": 2048,137        "n_mel": 128,138        "num_silent_frames": 8,139        "phase_dir": "phases",140        "phone_dir": "phones",141        "phone_energy_dir": "phone_energys",142        "phone_extractor": "espeak",143        "phone_pitch_dir": "phone_pitches",144        "phone_seq_file": "phone_seq_file",145        "pitch_bin": 256,146        "pitch_dir": "pitches",147        "pitch_extractor": "rmvpe",148        "pitch_max": 1100.0,149        "pitch_min": 50.0,150        "pitch_norm": false,151        "pitch_remove_outlier": false,152        "processed_dir": "data/",153        "raw_data": "raw_data",154        "real_dir": "reals",155        "ref_level_db": 20,156        "sample_rate": 44100,157        "spk2id": "singers.json",158        "symbols_dict": "symbols.dict",159        "train_file": "train.json",160        "trim_fft_size": 512,161        "trim_hop_size": 128,162        "trim_silence": false,163        "trim_top_db": 30,164        "trimmed_wav_dir": "trimmed_wavs",165        "use_amplitude_phase": false,166        "use_audio": true,167        "use_dur": false,168        "use_emoid": false,169        "use_frame_duration": false,170        "use_frame_energy": false,171        "use_frame_pitch": true,172        "use_lab": false,173        "use_label": false,174        "use_linear": false,175        "use_log_scale_energy": false,176        "use_log_scale_pitch": false,177        "use_mel": false,178        "use_min_max_norm_mel": false,179        "use_one_hot": false,180        "use_phn_seq": false,181        "use_phone": false,182        "use_phone_duration": false,183        "use_phone_energy": false,184        "use_phone_pitch": false,185        "use_spkid": false,186        "use_text": false,187        "use_uv": true,188        "use_wav": false,189        "use_wenet": false,190        "utt2emo": "utt2emo",191        "utt2spk": "utt2spk",192        "uv_dir": "uvs",193        "valid_file": "valid.json",194        "wav_dir": "wavs",195        "wenet_dir": "wenet",196        "win_size": 2048,197    },198    "supported_model_type": [199        "GANVocoder",200        "Fastspeech2",201        "DiffSVC",202        "Transformer",203        "EDM",204        "Autotune",205        "CD",206    ],207    "task_type": "afgen",208    "train": {209        "adamw": {210            "adam_b1": 0.8,211            "adam_b2": 0.99,212            "lr": 0.0001,213        },214        "batch_size": 16,215        "criterions": [216            "feature",217            "discriminator",218            "generator",219            "multimel",220            "codebook",221            "commitment",222        ],223        "dataloader": {224            "num_worker": 32,225            "persistent_workers": true,226            "pin_memory": true,227            "prefetch_factor": 4,228        },229        "ddp": true,230        "exponential_lr": {231            "lr_decay": 0.9999996,232        },233        "freeze_step": 0,234        "gradient_accumulation_step": 1,235        "keep_checkpoint_max": 5,236        "keep_last": [237            3,238            -1,239        ],240        "max_epoch": 1000000,241        "max_steps": 1000000,242        "multi_speaker_training": false,243        "optimizer": "AdamW",244        "random_seed": 114514,245        "reducelronplateau": {246            "factor": 0.8,247            "min_lr": 0.0001,248            "patience": 10,249        },250        "run_eval": [251            true,252        ],253        "sampler": {254            "drop_last": true,255            "holistic_shuffle": true,256        },257        "save_checkpoint_stride": [258            1,259        ],260        "save_checkpoints_steps": 10000,261        "save_summary_steps": 500,262        "scheduler": "ReduceLROnPlateau",263        "total_training_steps": 50000,264        "tracker": [265            "tensorboard",266        ],267        "valid_interval": 10000,268    },269    "use_custom_dataset": [],270}