I tried to adapt CMU Sphinx to my accen using the guide.
I'm unable to do keyword spotting with the this model with this command -
pocketsphinx_continuous \
-hmm /Desktop/cmusphinx/pocketsphinx-5prealpha/model/en-us-adapt/en-us-adapt/ \
-lm /Desktop/cmusphinx/pocketsphinx-5prealpha/model/en-us-adapt/en-us.lm.bin \
-dict /Desktop/cmusphinx/pocketsphinx-5prealpha/model/en-us-adapt/cmudict-en-us.dict \
-infile /Downloads/Record_0011Cleaned.wav \
-kws /Desktop/cmusphinx/KeyPhrase2.txt \
-logfn /dev/null \
-time yes > /tmp.txt
pocketsphinx_continuous
seems to be looking for all the words that are available to it in the dictionary and not just those defined in my keyphrase file. The output additionally includes tags like <s>
, /s>
and so on. Neither of these happen if I remove the hmm
, lm
, and dict
arguments.
Why is this happening and what can be done to fix it?
You can specify either language model search with -lm
argument or keyword search with -kws
argument, not them together. -lm
has higher priority and is used instead of -kws
in your case.
To solve the problem remove -lm
part from your command line.