author | Daniel Povey <dpovey@gmail.com> | |
Tue, 11 Oct 2016 21:05:55 +0000 (17:05 -0400) | ||
committer | GitHub <noreply@github.com> | |
Tue, 11 Oct 2016 21:05:55 +0000 (17:05 -0400) | ||
commit | a8de21fd76f2736d91aab30763a12646cb7c378b | |
tree | 7c193052b2e522f44e35b6028d3384e1e3ce6215 | tree | snapshot (tar.xz tar.gz zip) |
parent | 2eab95ae2bfa2950df6ca68d1e2926d7afdb01b7 | commit | diff |
Unk model (#1058)
* Some partial changes towards supporting unknown-word models based on a phone language model.
* Modifications to ARPA LM compilation to remove un-needed ARPA states (those that do nothing but back off).
* Adding previously omitted script egs/tedlium/s5_r2/local/run_unk_model.sh
* Adding previously omitted script utils/lang/internal/modify_unk_pron.py
* Fixes to egs/tedlium/s5_r2/local/run_unk_model.sh (thanks to @xiaohui-zhang for finding them).
* Some partial changes towards supporting unknown-word models based on a phone language model.
* Modifications to ARPA LM compilation to remove un-needed ARPA states (those that do nothing but back off).
* Adding previously omitted script egs/tedlium/s5_r2/local/run_unk_model.sh
* Adding previously omitted script utils/lang/internal/modify_unk_pron.py
* Fixes to egs/tedlium/s5_r2/local/run_unk_model.sh (thanks to @xiaohui-zhang for finding them).
21 files changed: