Skip to content

Conversation

@pegahgh
Copy link

@pegahgh pegahgh commented Feb 28, 2016

No description provided.

danpovey and others added 30 commits February 4, 2016 20:56
nnet2 versoin of steps/cleanup/find_bad_utts.sh
Implement the 3d maxpooling component.
…h-log

NormalizeComponent with add-log-std option
…component and statistics-pooling-component, including bug-fix to nnet-computation-graph.cc that only affects setups with general components.
…der of permutation; adding results (prior to script bug-fix); minor bug-fix in nnet-chain-example.cc which affects nothing
danpovey and others added 30 commits March 4, 2016 23:58
removed bugs introduced due automerge with master
…y learning-rate factor each time we add final xent layer
swbd : Added lstm chain recipe based on tdnn_6h
…0 to 5, to be the same as the code (turns out this inintended change was harmful to results, see run_tdnn_6z.sh)
…rity now that we have learning-rate-factors (the old interface was confusing and not powerful enough for all situations).
Minor changes required for nnet3 sequence training
swbd : Added blstm + chain recipe, results pending
add utils/perturb_data_signal_v2.sh
swbd+chain+lstm:updated results after Dan's fix to right-tolerance de…
We want to specify the language model weight range for decoding in
chain models, but previously lmrescore_const_arpa.sh did not allow
passing in arguments to score.sh. This is also necessary to specify
--decode-mbr false to score.sh for chain scoring in TEDLIUM.
…options

Add options for score.sh to rescoring script.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.