mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-26 18:24:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix incorrect doc.

Browse Source
This commit is contained in:
Fangjun Kuang 2021-11-15 12:18:16 +08:00
parent e3d7f21372
commit 57b9c8868b
1 changed files with 6 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
icefall/lm/rescore.py
View File

@ -39,8 +39,8 @@ The following shows the steps:
(iv) Use "b" as "src" and its shifted version as "tgt". (iv) Use "b" as "src" and its shifted version as "tgt".
We can get another likelihood value, denoted as "ab_other" We can get another likelihood value, denoted as "ab_other"
So for the path pair (a, b), (a, c), we can get the following log-likelihood So for the path pair (a, b), (a, c), (b, a), (b, c), (c, a), and (c, b),
values, viewed as two tensors: we can get the following log-likelihood values, viewed as two tensors:
self = [ab_self, ac_self, ba_self, bc_self, ca_self, cb_self] self = [ab_self, ac_self, ba_self, bc_self, ca_self, cb_self]
@ -219,7 +219,7 @@ def make_repeat_map(row_splits: torch.Tensor):
>>> row_splits = torch.tensor([0, 3, 5], dtype=torch.int32) >>> row_splits = torch.tensor([0, 3, 5], dtype=torch.int32)
>>> make_repeat_map(row_splits) >>> make_repeat_map(row_splits)
tensor([0, 1, 2, 0, 1, 2, 0, 1, 2, 3, 4, 3, 4], dtype=torch.int32) tensor([1, 2, 0, 2, 0, 1, 4, 3], dtype=torch.int32)
""" """
device = row_splits.device device = row_splits.device
@ -250,12 +250,11 @@ def make_repeat_map(row_splits: torch.Tensor):
def make_repeat(tokens: k2.RaggedTensor) -> k2.RaggedTensor: def make_repeat(tokens: k2.RaggedTensor) -> k2.RaggedTensor:
"""Repeat the number of paths of an utterance to the number that """Repeat paths in an utterance.
equals to the number of paths in the utterance.
For instance, if an utterance contains 3 paths: [path1 path2 path3], For instance, if an utterance contains 3 paths: [path1 path2 path3],
after repeating, this utterance will contain 9 paths: after repeating, this utterance will contain 6 paths:
[path1 path2 path3] [path1 path2 path3] [path1 path2 path3] [path2 path3] [path1 path3] [path1 path2]
>>> tokens = k2.RaggedTensor([ [[1, 2, 3], [4, 5], [9]], [[5, 8], [10, 1]] ]) >>> tokens = k2.RaggedTensor([ [[1, 2, 3], [4, 5], [9]], [[5, 8], [10, 1]] ])
>>> tokens.to_str_simple() >>> tokens.to_str_simple()