Skip to content
Snippets Groups Projects
Commit 5a0b370f authored by tanel's avatar tanel
Browse files

Added adaptation-state property, needs fairly recent Kaldi to compile

parent 5f5ef055
Branches
No related tags found
No related merge requests found
Show whitespace changes
Inline Side-by-side
Showing
with 10 additions and 2 deletions
README.md
+ 10
− 2
View file @ 5a0b370f
...@@ -4,10 +4,18 @@ ...@@ -4,10 +4,18 @@
GStreamer plugin that wraps Kaldi's SingleUtteranceNnet2Decoder. It requires iVector-adapted GStreamer plugin that wraps Kaldi's SingleUtteranceNnet2Decoder. It requires iVector-adapted
DNN acoustic models. The iVectors are adapted to the current audio stream automatically. DNN acoustic models. The iVectors are adapted to the current audio stream automatically.
The iVectors are reset after the decoding session (stream) ends. ~~The iVectors are reset after the decoding session (stream) ends.
Currently, it's not possible to save the adaptation state and recall it later 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 for a particular speaker, to make the adaptation persistent over multiple decoding
sessions. sessions.~~
Update: 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).
Use the `adaptation-state` plugin property to get, and set the adaptation state. Use an empty string
with the set method to reset the adaptation state to default. This functionality requires
Kaldi revision 4582 or later.
# HOW TO COMPILE IT # HOW TO COMPILE IT
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment