bandit-v2/dnr-v3-com-smad-multi-v2b.yaml

datasets:
  train:
    cls: DivideAndRemasterRandomChunkDataset
    kwargs:
      split: train
      subset: ${....commons.datasets.subset}
      stems: ${....commons.datasets.stems}
      npy_memmap: ${....commons.datasets.npy_memmap}
      full_audio_length_seconds: ${....commons.datasets.full_audio_length_seconds}
      data_root: ${....commons.datasets.data_root}
      chunk_size_seconds: ${....commons.datasets.chunk_size_seconds}
      target_dataset_length: 16384 #10240
      aligned_chunking: false
      auto_include_mixture: true
      recompute_mixture: true

  val:
    cls: DivideAndRemasterDeterministicChunkDataset
    kwargs:
      split: val
      subset: ${....commons.datasets.subset}
      stems: ${....commons.datasets.stems}
      npy_memmap: ${....commons.datasets.npy_memmap}
      full_audio_length_seconds: ${....commons.datasets.full_audio_length_seconds}
      data_root: ${....commons.datasets.data_root}
      chunk_size_seconds: ${....commons.datasets.chunk_size_seconds}
      hop_size_seconds: ${....commons.datasets.hop_size_seconds}
      aligned_chunking: true
      auto_include_mixture: true
      recompute_mixture: true
  test:
    cls: DivideAndRemasterFullTrackDataset
    kwargs:
      split: test
      subset: ${....commons.datasets.subset}
      stems: ${....commons.datasets.stems}
      npy_memmap: ${....commons.datasets.npy_memmap}
      full_audio_length_seconds: ${....commons.datasets.full_audio_length_seconds}
      data_root: ${....commons.datasets.data_root}


commons:
  datasets:
    subset: multi-v2b
    stems:
      - speech
      - music
      - sfx
    npy_memmap: true
    full_audio_length_seconds: 60.0
    chunk_size_seconds: 8.0
    hop_size_seconds: 8.0
    data_root: /fsx_vfx/projects/aa-cass/data/dnr-variants/dnr-v3-com-48k-smad-multi-v2

datamodule:
  cls: DivideAndRemasterDataModule
  kwargs:
    batch_size_per_gpu: 8
    num_workers: 8