accumulate_grad_batches: 1 acous_params: - - 480 - 1200 - 80 - - 240 - 1200 - 160 adam_b1: 0.8 adam_b2: 0.99 amp: true audio_num_mel_bins: 160 audio_sample_rate: 44100 base_config: - egs/foa_vae/wavvae_v3.yaml batch_by_size: true binary_data_dir: data/binary_v3/sa_data_v1 check_val_every_n_epoch: 10 ckpt_milestone_interval: 10000 clip_grad_norm: 1.0 clip_grad_value: 0 d_pretrain_steps: 10000 dataloader_mp_type: fork dataloader_version: v2 dataset_cls: tasks.foa_vae.dataset_utils.pyroom_dataset.FOAVAEV1Dataset datasets: pyroom_foa: processer_fn: tasks.foa_vae.dataset_utils.pyroom_dataset.processer_fn_pyroom read_idx: false reader_chunk_size: 8 train_sets: - /mnt/bn/sa-ag-data/leike/spatial_edit/triplet/metadata_training/audio_edit_train.jsonl - /mnt/bn/sa-ag-data/leike/spatial_edit/triplet2/metadata_training2/audio_edit_train.jsonl - /mnt/bn/sa-ag-data/leike/spatial_edit/triplet2/metadata_training2/joint_edit_train.jsonl weight: 0.5 real_foa: processer_fn: tasks.foa_vae.dataset_utils.pyroom_dataset.processer_fn_pyroom read_idx: false reader_chunk_size: 8 train_sets: - /mnt/bn/sa-ag-data/leike/spatial/data/metadata/yt-ambien_trval_vae.jsonl - /mnt/bn/sa-ag-data/leike/spatial/data/metadata/sphere360_vae.jsonl weight: 1.0 spat_libri: processer_fn: tasks.foa_vae.dataset_utils.pyroom_dataset.processer_fn_pyroom read_idx: false reader_chunk_size: 8 train_sets: - /mnt/bn/sa-ag-data/leike/spatial/data/metadata/spat_libri.jsonl weight: 1.0 ddp_dir: '' debug: false disc_lr: 1.0e-05 disc_weight_decay: 0.0 discriminator_grad_norm: 1.0 dropout: 0.0 ds_workers: 8 dtype_save_bf16: true dynamic_batch: true ema_decay: 0.9999 ema_power: 0.75 ema_update_after_step: 1 ema_use_warmup: true endless_ds: true eval_max_batches: 0 exp_name: 260720_foa_vaev3_post f0_max: 600 f0_min: 60 fast_ds: true fast_ds_prefetch_steps: 8 fast_ds_shuffle_buffer: 4 fft_size: 1200 find_unused_parameters: true fmax: 12000 fmin: 0 foa_align_to_frames_multiple: true foa_audio_roots: [] foa_bucket_len: 35 foa_channel_order: WYZX foa_expected_channels: 4 foa_normalize_mode: none foa_path_fields: - target_wav_path - wav_path - foa_path - audio_path - target_path - path - file_path foa_projection: gain: 1.0 include_axes: true include_corners: false mrstft_chunk_size: 2 random_dirs: 0 foa_random_crop: true foa_spatial_loss: energy_percentile: 0.5 fft_size: 1024 hop_size: 256 win_length: 1024 foa_target_seconds: 1.486077097505669 foa_vae: downsampling_ratios: - 2 - 2 - 4 - 8 - 8 input_channels: 4 latent_channels: 64 output_channels: 4 pretrained_model_dir: checkpoints/vae frames_multiple: 8 from_scratch: false gan_ramp_steps: 20000 generator_grad_norm: 1.0 grad_norm_skip_init_threshold: 50000.0 grad_norm_skip_threshold: 50000.0 grad_norm_skip_warmup_steps: 100 hop_size: 240 infer: false infer_save_dir: '' init_method: tcp load_ckpt: '' load_ckpt_strict: false load_map_to_gpu: false load_wav: true log_grad_every_n_steps: 1 loss_configs: discriminator: config: filters: 32 hop_lengths: - 512 - 256 - 128 - 64 - 32 n_ffts: - 2048 - 1024 - 512 - 256 - 128 win_lengths: - 2048 - 1024 - 512 - 256 - 128 type: encodec high_frequency_excess_db: fft_size: 2048 hop_size: 512 margin_db: 1.0 min_db: -80.0 min_frequency_hz: 6000.0 win_length: 2048 mel_multiband: band_edges_hz: - 0 - 1000 - 6000 - 20000 band_weights: - 1.0 - 1.0 - 1.0 resolutions: - audio_num_mel_bins: 80 fft_size: 512 fmax: 20000 fmin: 0 hop_size: 128 win_size: 512 - audio_num_mel_bins: 128 fft_size: 1024 fmax: 20000 fmin: 0 hop_size: 256 win_size: 1024 - audio_num_mel_bins: 160 fft_size: 2048 fmax: 20000 fmin: 0 hop_size: 512 win_size: 2048 spectral: config: fft_sizes: - 2048 - 1024 - 512 - 256 - 128 - 64 - 32 hop_sizes: - 512 - 256 - 128 - 64 - 32 - 16 - 8 perceptual_weighting: true win_lengths: - 2048 - 1024 - 512 - 256 - 128 - 64 - 32 losses: lambda_adv: 0.05 lambda_cov: 1.0 lambda_cross_phase: 0.5 lambda_dis: 1.0 lambda_feature_matching: 5.0 lambda_high_frequency_excess_db: 0.01 lambda_intensity_dir: 1.0 lambda_kl: 1.0e-05 lambda_mel_multiband: 0.0 lambda_mrstft: 1.0 lambda_proj_mrstft: 1.0 loud_norm: false lr: 1.0e-05 master_addr: '' max_epochs: 1000 max_frames: 12000 max_sentences: 20 max_tokens: 300 max_updates: 1500000 max_valid_sentences: 1 max_valid_tokens: 60000 mel_vmax: 0.5 mel_vmin: -6 min_frames: 50 mixed_precision: bf16 mmap: true num_ckpt_keep: 5 num_sanity_val_steps: 10 num_workers: 16 optimizer: betas: - 0.9 - 0.999 fused: false lr: 7.5e-05 weight_decay: 0.01 prefetch_factor: 2 print_nan_grads: true reduce_transient_noise: false rename: -1 rename_tmux: false resume_from: checkpoints/260719_foa_vaev3_sphere_fixed resume_from_checkpoint: 0 sample_rate: 44100 sample_size: 65536 save_best: false save_codes: - tasks - modules - egs - utils scheduler: cosin_annealing scheduler_initial_period: 10000 scheduler_lr_min: 1.0e-05 scheduler_period_mult: 1.2 seed: 42 sort_by_len: true sp_size: 1 start_rank: -1 task_cls: tasks.foa_vae.wavvae_v3_post_task.FOAWavVAEV3PostTask tb_log_interval: 100 test_max_batches: 100 test_num: 100 test_set_name: test tgt_size_max: 200 tgt_size_min: 20 torch_compile: false train_set_name: train use_dataset_weight: true use_ema: true use_fast_dataloader: true use_fsdp: false vae_stride: 4 val_batch_size: 0 val_check_interval: 2000 val_dataset_cls: tasks.foa_vae.dataset_utils.pyroom_dataset.FOAVAEV1Dataset val_max_batches: 100 valid_monitor_key: val_loss valid_monitor_mode: min valid_set_name: valid validate: false warmup_updates: 2000 weight_decay: 0.0 win_size: 1200 work_dir: checkpoints/260720_foa_vaev3_post world_size: -1