text_query/config.yaml

DASM_HTSAT:
  init_kwargs:
    backbone_param:
      embed_dim: 768
      pretrain_model_path: ./pretrained_model/detect_any_sound/htast_mga.pt
      lora_config: 
    cnn_param:
      n_in_channel: 1
      activation: cg
      conv_dropout: 0.5
      kernel_size: [3, 3, 3, 3, 3, 3, 3, 3, 3, 3]
      padding: [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
      stride: [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
      nb_filters: [ 16, 16, 32, 32, 64, 64, 128, 128, 256, 256]
      pooling: [ [ 2, 2 ], [1, 1], [ 2, 2 ],[1, 1], [ 1, 2 ], [ 1, 2], [ 1, 2 ], [ 1, 1 ], [ 1, 2], [1, 1] ]
    at_param:
      at_decoder_layer: 2
      query_projector: True
      out_type: "sigmoid"
      query_dim: 1024
      query: ./pretrained_model/detect_any_sound/text_query/text_query.pt

    backbone_upsample_ratio: 15.626                     # decode ratio in local gru decoder
    decoder_dim: 384
    num_heads: 12
    decoder: "conformer"
    decoder_layer_num: 2
    decoder_pos_emd_len: 500
    decoder_expand_rate: 1
    class_num: 407

  train_kwargs:
    # windows parameters
    encoder_win: False    # constant
    # temperature parameters
    temp_w: 1
    
  val_kwargs:
    # windows parameters
    encoder_win: False
    temp_w: 0.5

  test_kwargs:
    # windows parameters
    encoder_win: False
    temp_w: 0.5


feature:                               
  pred_len: 500
  n_mels: 64
  n_fft: 1024
  hopsize: 320
  win_length: 1024
  fmin: 50
  fmax: 14000
  audio_max_len: 10
  sr: 32000
  net_subsample: 2