hyperparams.yaml

# ################################
# Model: Pretrained RE-SepFormer for speech separation
# Dataset : WSJ0-2mix 
# ################################

sample_rate: 8000
num_spks: 2

# Encoder parameters
N_encoder_out: 128
out_channels: 128
kernel_size: 16
kernel_stride: 8

# Specifying the network
Encoder: !new:speechbrain.lobes.models.dual_path.Encoder
  kernel_size: 16
  out_channels: 128

intra_mdl: !new:speechbrain.lobes.models.resepformer.SBTransformerBlock_wnormandskip
  num_layers: 8
  d_model: 128
  nhead: 8
  d_ffn: 1024
  dropout: 0
  use_positional_encoding: true
  norm_before: true
  use_norm: true
  use_skip: true

mem_mdl: !new:speechbrain.lobes.models.resepformer.SBTransformerBlock_wnormandskip
  num_layers: 8
  d_model: 128
  nhead: 8
  d_ffn: 1024
  dropout: 0
  use_positional_encoding: true
  norm_before: true
  use_norm: true
  use_skip: true

MaskNet: !new:speechbrain.lobes.models.resepformer.ResourceEfficientSeparator
  input_dim: 128
  num_spk: 2
  causal: false
  unit: 256
  segment_size: 150
  layer: 2
  mem_type: av
  seg_model: !ref <intra_mdl>
  mem_model: !ref <mem_mdl>
  
Decoder: !new:speechbrain.lobes.models.dual_path.Decoder
  in_channels: 128
  out_channels: 1
  kernel_size: 16
  stride: 8
  bias: false

modules:
  encoder: !ref <Encoder>
  decoder: !ref <Decoder>
  masknet: !ref <MaskNet>

pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
  loadables:
     encoder: !ref <Encoder>
     masknet: !ref <MaskNet>
     decoder: !ref <Decoder>