Commit 864b6b3a authored by tanel's avatar tanel
Browse files

Implemented optional rescoring with (large) a 'constant ARPA' LM

parent 3be2d7ac
......@@ -4,12 +4,18 @@
GStreamer plugin that wraps Kaldi's SingleUtteranceNnet2Decoder. It requires iVector-adapted
DNN acoustic models. The iVectors are adapted to the current audio stream automatically.
~~The iVectors are reset after the decoding session (stream) ends.
Currently, it's not possible to save the adaptation state and recall it later
for a particular speaker, to make the adaptation persistent over multiple decoding
sessions.~~
Update: the plugin saves the adaptation state between silence-segmented utterances and between
# CHANGELOG
2015-01-09: Added language model rescoring functionality. In order to use it,
you have to specify two properties: `lm-fst` and `big-lm-const-arpa`. The `lm-fst`
property gives the location of the *original* LM (the one that was used fpr
compiling the HCLG.fst used during decodong). The `big-lm-const-arpa` property
gives the location of the big LM used that is used to rescore the final lattices.
The big LM must be in the 'ConstArpaLm' format, use the Kaldi's
`utils/build_const_arpa_lm.sh` script to produce it from the ARPA format.
2014-11-11: the plugin saves the adaptation state between silence-segmented utterances and between
multiple decoding sessions of the same plugin instance.
That is, if you start decoding a new stream, the adaptation state of the
previous stream is used (unless it's the first stream, in which case a global mean is used).
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment