mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-09 01:52:41 +00:00
Code Issues Packages Projects Releases Wiki Activity

Small fix for frame_reducer.py (#871)

Browse Source
This commit is contained in:
Yifan Yang 2023-02-03 17:49:54 +08:00 committed by GitHub
parent bffce413f0
commit 029c8566e4
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
egs/librispeech/ASR/pruned_transducer_stateless7_ctc_bs/frame_reducer.py
View File

@ -44,6 +44,7 @@ class FrameReducer(nn.Module):
x: torch.Tensor, x: torch.Tensor,
x_lens: torch.Tensor, x_lens: torch.Tensor,
ctc_output: torch.Tensor, ctc_output: torch.Tensor,
blank_id: int = 0,
) -> Tuple[torch.Tensor, torch.Tensor]: ) -> Tuple[torch.Tensor, torch.Tensor]:
""" """
Args: Args:
@ -54,6 +55,8 @@ class FrameReducer(nn.Module):
`x` before padding. `x` before padding.
ctc_output: ctc_output:
The CTC output with shape [N, T, vocab_size]. The CTC output with shape [N, T, vocab_size].
blank_id:
The blank id of ctc_output.
Returns: Returns:
out: out:
The frame reduced encoder output with shape [N, T', C]. The frame reduced encoder output with shape [N, T', C].
@ -65,7 +68,7 @@ class FrameReducer(nn.Module):
N, T, C = x.size() N, T, C = x.size()
padding_mask = make_pad_mask(x_lens) padding_mask = make_pad_mask(x_lens)
non_blank_mask = (ctc_output[:, :, 0] < math.log(0.9)) * (~padding_mask) non_blank_mask = (ctc_output[:, :, blank_id] < math.log(0.9)) * (~padding_mask)
out_lens = non_blank_mask.sum(dim=1) out_lens = non_blank_mask.sum(dim=1)
max_len = out_lens.max() max_len = out_lens.max()