Skip to content
This repository has been archived by the owner on Oct 13, 2022. It is now read-only.

[WIP] 2-state HMM topo as an alternative to CTC topo #126

Open
wants to merge 12 commits into
base: master
Choose a base branch
from
Open

[WIP] 2-state HMM topo as an alternative to CTC topo #126

Changes from 1 commit
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
8586e31
Add option to build 2-state HMM topology for MMI graphs
pzelasko Mar 12, 2021
2221209
Use the 2-state HMM topo in MMI conformer script
pzelasko Mar 12, 2021
d2696ca
Fix HMM topo sorting and followup tokens
pzelasko Mar 13, 2021
d1efefa
Adjust MMI training script and decrease beam for dens
pzelasko Mar 13, 2021
1f35b96
Fix number of nnet outputs and decrease beam
pzelasko Mar 13, 2021
ee992eb
Decoding changes for 2-state HMM topo
pzelasko Mar 15, 2021
497e075
Merge remote-tracking branch 'origin/feature/hmm-2state-topo' into fe…
pzelasko Mar 15, 2021
5eebbac
Beam/dir changes
pzelasko Mar 15, 2021
268898d
Merge remote-tracking branch 'upstream/master'
pzelasko Mar 16, 2021
bdc812e
Merge branch 'master' into feature/hmm-2state-topo
pzelasko Mar 16, 2021
eac0a65
Fix possible issue with incompatible HLG being loaded
pzelasko Mar 16, 2021
c85e53a
fix for HLG loading
pzelasko Mar 16, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Fix HMM topo sorting and followup tokens
  • Loading branch information
@pzelasko
pzelasko committed Mar 13, 2021
commit d2696ca9f4d18ed52408a773118c46a6e2730c00
8 changes: 6 additions & 2 deletions snowfall/training/hmm_topo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,11 @@ def build_hmm_topo_2state(tokens: List[int]) -> k2.Fsa:
Returns:
An FST that converts a sequence of HMM state IDs to a sequence of token IDs.
"""
followup_tokens = range(len(tokens), len(tokens) * 2)
min_token_id = min(tokens)
followup_tokens = list(range(
len(tokens) + min_token_id,
2 * len(tokens) + min_token_id
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

are you making an assumption here that tokens is contiguous?

))
num_states = len(tokens) + 2 # + start state, + final state
arcs = []

Expand All @@ -42,7 +46,7 @@ def build_hmm_topo_2state(tokens: List[int]) -> k2.Fsa:
arcs += [f'{num_states - 1}']

# Build the FST
arcs = '\n'.join(sorted(arcs))
arcs = '\n'.join(sorted(arcs, key=lambda arc: int(arc.split()[0])))
ans = k2.Fsa.from_str(arcs)
ans = k2.arc_sort(ans)
return ans