| data: | |
| block_size: 512 | |
| dataset_path: ../datasets/vctk-partial | |
| duration: 1.8 | |
| encoder: dpwavlmbase | |
| encoder_ckpt: models/pretrained/dphubert/DPWavLM-sp0.75.pth | |
| encoder_hop_size: 320 | |
| encoder_out_channels: 768 | |
| encoder_sample_rate: 16000 | |
| extensions: | |
| - wav | |
| f0_extractor: rmvpe | |
| f0_max: 1200 | |
| f0_min: 65 | |
| sampling_rate: 44100 | |
| spk_embed_channels: 256 | |
| spk_embed_encoder: pyannote.audio | |
| spk_embed_encoder_ckpt: ./models/pretrained/pyannote.audio/wespeaker-voxceleb-resnet34-LM/pytorch_model.bin | |
| spk_embed_encoder_sample_rate: 16000 | |
| volume_window_size: 8 | |
| device: cuda | |
| env: | |
| expdir: ../datasets/exp/vctk-partial | |
| gpu_id: 0 | |
| loss: | |
| beta: 0.8 | |
| fft_max: 2048 | |
| fft_min: 256 | |
| n_scale: 4 | |
| overlap: 0.5 | |
| use_dual_scale: false | |
| use_dual_scale_log_freq: true | |
| model: | |
| f0_input_variance: 0.0 | |
| f0_offset_size_downsamples: 8 | |
| harmonic_env_size_downsamples: 8 | |
| no_use_embed_conv: false | |
| noise_env_size_downsamples: 8 | |
| noise_seed: 289 | |
| noise_to_harmonic_phase: true | |
| type: CombSubMinimumNoisedPhase | |
| units_hidden_channels: 256 | |
| units_layers: | |
| - - 10 | |
| - 11 | |
| use_f0_offset: true | |
| use_harmonic_env: false | |
| use_noise_env: true | |
| use_speaker_embed: true | |
| win_length: 2048 | |
| train: | |
| amp_dtype: fp32 | |
| batch_size: 48 | |
| cache_all_data: true | |
| cache_device: cuda | |
| cache_fp16: true | |
| epochs: 50000 | |
| frame_hop_random_max: 64 | |
| frame_hop_random_min: 32 | |
| interval_log: 10 | |
| interval_val: 2000 | |
| loss_variation: 0.1 | |
| low_similar_loss_variation: 0.7 | |
| lr: 0.0005 | |
| num_workers: 2 | |
| only_u2c_stack: false | |
| save_opt: false | |
| sched_cooldown: 2 | |
| sched_factor: 0.5 | |
| sched_min_lr: 3.0e-06 | |
| sched_patience: 30 | |
| sched_threshold: 1.0e-05 | |
| sched_threshold_mode: rel | |
| weight_decay: 0 | |