espnet2.legacy.nets.pytorch_backend.transducer.utils.select_lm_state

source

espnet2.legacy.nets.pytorch_backend.transducer.utils.select_lm_state(lm_states: List[Any] | Dict[str, Any], idx: int, lm_layers: int, is_wordlm: bool) → List[Any] | Dict[str, Any]

Get ID state from LM hidden states.

• Parameters:
  • lm_states – LM hidden states.
  • idx – LM state ID to extract.
  • lm_layers – Number of LM layers.
  • is_wordlm – Whether provided LM is a word-level LM.
• Returns: LM hidden state for given ID.
• Return type: idx_state