Speech Recognition Engines

Flat
RE: Julius Speech Recognition
User: Gururaj
Date: 1/12/2017 6:10 am
Views: 7526
Rating: 0

Hi all,

I am using julius speech recognition for my application,  i want to know how to control the julius...i am using live microphone to give the inputs to julius, the problem i am facing is...whenever i give input through the live microphone, for example: the command is play song and if this is not present in the .dict file, even though julius will return some random command which is present in the .dict file...i want to know how to control this and  make  it more flexible..like i want julius to return nothing or search failed command as output if the given input is not present in the .dict file.

can anyone guide me in this....thank

--- (Edited on 1/12/2017 6:10 am [GMT-0600] by Gururaj) ---

RE: Julius Speech Recognition
User: kmaclean
Date: 1/12/2017 9:00 am
Views: 1387
Rating: 0

Hi Gururaj,

You can look at the confidence score returned by Julius to only allow recognition results that are higher than some minimum threshold.  For voxshell I used a minimum confidence of 0.7 (i.e. int min_confidence=.7; ) - The Simon dialog manager uses a similar approach.

Ken

 

--- (Edited on 1/12/2017 10:00 am [GMT-0500] by kmaclean) ---

RE: Julius Speech Recognition
User: Gururaj
Date: 1/17/2017 6:01 am
Views: 56
Rating: 0

thanks for the reply...but when i use min_confidence also the same thing repeats...it will identify random words only..

--- (Edited on 1/17/2017 6:01 am [GMT-0600] by Gururaj) ---

RE: Julius Speech Recognition
User: colbec
Date: 1/18/2017 5:21 am
Views: 292
Rating: 0

Julius has a very narrow view of the world. It can only recognize what it knows from the grammar. Julius is doing its job correctly by only returning the closest match that is valid from the grammar. It may appear to be random but that's not the case, Julius is giving you the closest match.

You can check what Julius thinks about "play sound" given your grammar using the utility accept_check like:

> accept_check mygrammar
Stat: init_voca: read 53 words
Reading in term file (optional)...done
9 categories, 53 words
DFA has 15 nodes and 28 arcs
-----
please input word sequence><s> three four </s>
wseq: <s> three four </s>
cate: NS_B DIGITZ DIGITZ NS_E
accepted
please input word sequence><s> play sound </s>
rejected at 2: word "play" not in voca

In this case Julius knows about three four so it is accepted, otherwise it would be rejected.

You might need to play with the value of your threshhold to get it to reject "play sound" if it is not in the grammar. Run it a few times and see what the confidence value is, and revise your threshhold on the basis of what Julius is doing.

--- (Edited on 2017-01-18 6:21 am [GMT-0500] by colbec) ---

RE: Julius Speech Recognition
User: Gururaj
Date: 3/1/2017 12:09 am
Views: 2411
Rating: 0

Hi i am still not able to completely control the julius, can you please guide me further...what extra i need to do for this. Thank

--- (Edited on 3/1/2017 12:09 am [GMT-0600] by Gururaj) ---

--- (Edited on 3/1/2017 12:09 am [GMT-0600] by Gururaj) ---

PreviousNext