espnet2.asr_transducer.beam_search_transducer.ExtendedHypothesis

source

class espnet2.asr_transducer.beam_search_transducer.ExtendedHypothesis(score: float, yseq: List[int], dec_state: Tuple[Tensor, Tensor | None] | None = None, lm_state: Dict[str, Any] | List[Any] | None = None, dec_out: Tensor | None = None, lm_score: Tensor | None = None)

Bases: Hypothesis

Extended hypothesis definition for NSC beam search and mAES.

This class extends the default hypothesis to include additional attributes specifically useful for the NSC beam search and modified Adaptive Expansion Search (mAES) algorithms.

dec_out

Decoder output sequence. Shape: (B, D_dec)

• Type: torch.Tensor

lm_score

Log-probabilities of the language model for the given label. Shape: (vocab_size)

• Type: torch.Tensor

• Parameters:

  • score – Total log-probability.
  • yseq – Label sequence as integer ID sequence.
  • dec_state – RNN/MEGA Decoder state (None if Stateless).
  • lm_state – RNNLM state. ((N, D_lm), (N, D_lm)) or None.
  • dec_out – Decoder output sequence. Shape: (B, D_dec).
  • lm_score – Log-probabilities of the LM for given label. Shape: (vocab_size).

Examples

>>> hyp = ExtendedHypothesis(
...     score=-1.0,
...     yseq=[1, 2, 3],
...     dec_state=(torch.tensor([0.5]), None),
...     lm_state=None,
...     dec_out=torch.tensor([[0.1, 0.2], [0.3, 0.4]]),
...     lm_score=torch.tensor([0.5, 0.6])
... )
>>> print(hyp.score)
-1.0
>>> print(hyp.yseq)
[1, 2, 3]

dec_out

*: Tensor* *= None*

lm_score

*: Tensor* *= None*