The outputs of keyword search and ngram search are attached - they are plain text files
Please advise whether there is anything specific that needs to be done for training a keyword search model?
EDIT: Just to clarify, the Keyword search using the supplied US-EN model is working. So is it something specific to the training mode settings or configuration parameters?
Many thanks for your prompt reply. I can live with Grammar search and implement my own keyword spotting with my small vocabulary.
Can a jsgf file be used instead of a *.lm in the config file for training
$CFG_LANGUAGEMODEL = "$CFG_LIST_DIR/$CFG_DB_NAME.lm";
It is not explicitly stated in the tutorial.
Secondly if it is allowed if there are two rules in the jsgf file will it use both the rules?
The reason why I am asking this is, I just ran pocketsphinx_continuous detection using a jsgf file which had two rules
<wakeup> <command1></command1></wakeup>
the detector only picked the first occuring rule. Is this how it is supposed to work?
Last edit: Q3Varnam 2018-04-11
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have created a own language model - non english. I have trained it with around 30 minutes of data. Its a small vocabulary model.
When I run PocketSphinx continuous in ngram search mode - it detects the words spoken from the mic with 99% accuracy.
However when I try to run it in the Keyword search mode - nothing gets detected at all.
I have tested it with a recorded wav file to avoid any issues with ambient noise etc.
Even with the wav file, ngram search detects it where as keyword search mode fails.
My keyworld list file (kwlist) has
The outputs of keyword search and ngram search are attached - they are plain text files
Please advise whether there is anything specific that needs to be done for training a keyword search model?
EDIT: Just to clarify, the Keyword search using the supplied US-EN model is working. So is it something specific to the training mode settings or configuration parameters?
Last edit: Q3Varnam 2018-04-11
This dataset size is way below our recommendation for model training.
keyword spotting requires large vocabulary model.
Many thanks for your prompt reply. I can live with Grammar search and implement my own keyword spotting with my small vocabulary.
Can a jsgf file be used instead of a *.lm in the config file for training
$CFG_LANGUAGEMODEL = "$CFG_LIST_DIR/$CFG_DB_NAME.lm";
It is not explicitly stated in the tutorial.
Secondly if it is allowed if there are two rules in the jsgf file will it use both the rules?
The reason why I am asking this is, I just ran pocketsphinx_continuous detection using a jsgf file which had two rules
<wakeup> <command1></command1></wakeup>
the detector only picked the first occuring rule. Is this how it is supposed to work?
Last edit: Q3Varnam 2018-04-11