torchaudio.models.squim_objective_model

torchaudio.models.squim_objective_model(feat_dim: int, win_len: int, d_model: int, nhead: int, hidden_dim: int, num_blocks: int, rnn_type: str, chunk_size: int, chunk_stride: Optional[int] = None) → SquimObjective

Build a custome torchaudio.prototype.models.SquimObjective model.

Parameters:

• feat_dim (int, optional) – The feature dimension after Encoder module.

• win_len (int) – Kernel size in the Encoder module.

• d_model (int) – The number of expected features in the input.

• nhead (int) – Number of heads in the multi-head attention model.

• hidden_dim (int) – Hidden dimension in the RNN layer of DPRNN.

• num_blocks (int) – Number of DPRNN layers.

• rnn_type (str) – Type of RNN in DPRNN. Valid options are [“RNN”, “LSTM”, “GRU”].

• chunk_size (int) – Chunk size of input for DPRNN.

• chunk_stride (int or None, optional) – Stride of chunk input for DPRNN.