| data: | |
| sampling_rate: 44100 | |
| block_size: 512 | |
| duration: 2 | |
| encoder: vec768l12 | |
| cnhubertsoft_gate: 10 | |
| encoder_sample_rate: 16000 | |
| encoder_hop_size: 320 | |
| encoder_out_channels: 768 | |
| training_files: filelists/train.txt | |
| validation_files: filelists/val.txt | |
| extensions: | |
| - wav | |
| unit_interpolate_mode: nearest | |
| model: | |
| type: Diffusion | |
| n_layers: 20 | |
| n_chans: 512 | |
| n_hidden: 256 | |
| use_pitch_aug: true | |
| timesteps: 1000 | |
| k_step_max: 0 | |
| n_spk: 1 | |
| device: cuda | |
| vocoder: | |
| type: nsf-hifigan | |
| ckpt: pretrain/nsf_hifigan/model | |
| infer: | |
| speedup: 10 | |
| method: dpm-solver++ | |
| env: | |
| expdir: logs/44k/diffusion | |
| gpu_id: 0 | |
| train: | |
| num_workers: 4 | |
| amp_dtype: fp32 | |
| batch_size: 48 | |
| cache_all_data: true | |
| cache_device: cpu | |
| cache_fp16: true | |
| epochs: 100000 | |
| interval_log: 10 | |
| interval_val: 2000 | |
| interval_force_save: 5000 | |
| lr: 0.0001 | |
| decay_step: 100000 | |
| gamma: 0.5 | |
| weight_decay: 0 | |
| save_opt: false | |
| spk: | |
| nanami: 0 | |