pocketsphinx - recognize few words

[Alfredo]

Hi all!

I have a small dictionary (like 6/7 words) for my project, with words like "go", "stop", "turn left", "turn right". So far i managed to create the dictionary and language model using lmtool. I've tested it in live speech mode and recognized all words without any problems.
The problem comes when i say a word that are not in the dictionary. For example if i say "car"(that are no in the dictonary) sphinx will return some other word that match to.
So, there is any solution for this? There is any way to just return a word if sphinx has like 80%(or something like that) sure that is the correct word that was spelled?

Thanks in advance!

[Arseniy Gorin]

For your task it is better to use keyword phrase list instead of LM http://cmusphinx.sourceforge.net.minilinx.com/wiki/tutoriallm

[Alfredo]

Thanks for reply!

As you suggested, using keyword phrase lead me up (in tutorial) to something like this:

oh mighty computer /1e-40/
hello world /1e-30/
other phrase /1e-20/

1. What does "1e-40" means?
2. The using of keyword phrase is the same as dictionary in code? I mean, in my code i have the path to where language model is, so in case i want to use a keyword phrase do i just need to change that path to the file with the keywords phrases?

[Arseniy Gorin]

What does "1e-40" means?

It's threshold that you can tune for each word or phrase

The using of keyword phrase is the same as dictionary in code? I mean, in my code i have the path to where language model is, so in case i want to use a keyword phrase do i just need to change that path to the file with the keywords phrases?

You still need the dictionary. Keyword list replaces language model

[Alfredo]

I've said it wrong...nvm

I forget to say Im using python for this. So far i got something like this

speech = LiveSpeech(
    verbose=False,
    sampling_rate=16000,
    buffer_size=2048,
    no_search=False,
    full_utt=False,
    hmm= os.path.join(model_path,'en-us'),
    lm=os.path.join(model_path, 'drone.lm.bin'),
    dic=os.path.join(model_path, 'drone.dict')
)

which is working fine. My doubt is how to use keyword list instead of language model.
The keyword list is a .txt ? or .list? And then if i have a file with the correct extension, how do i use it in code? Should i do something like "kws = os.path.join(model_path,keyword)" ? Can you describe the steps to change from language model to keyword list?

[Arseniy Gorin]

You can find an example with kws phrase here
https://github.com/cmusphinx/pocketsphinx/blob/master/swig/python/test/kws_test.py

You should replace the phrase with a list. Yes, it is a text file with phrases and thresholds as described in LM tutorial