Click here to register.

Speech Recognition Engines

Flat
Getting started for dummies
User: Rubiksmomo
Date: 10/2/2016 12:33 pm
Views: 2566
Rating: 0

Hey. I'm trying to get speech recognition working on Arch Linux ARM on Raspberry Pi 3 model B. I've managed to get Julius running with the sample config. It actually recognizes some of the sample words sometimes if I speak close the mic. I am however overwhelmed by these long and confusing tutorials. Is there any easy way to teach Julius a single english word for starters? Or would you recommend some other software? If so, how would I set it up?

--- (Edited on 10/2/2016 12:33 pm [GMT-0500] by ) ---

Re: Getting started for dummies
User: kmaclean
Date: 10/4/2016 8:33 am
Views: 73
Rating: 0

>I am however overwhelmed by these long and confusing tutorials.

The tutorials are to create your own acoustic model.  You can use a current one from the nightly builds.

>Is there any easy way to teach Julius a single english word for starters?

Best to use more than a single word.

All you need is to create and compile your own grammar file, which create a .dict and .dfa file, then update your julius config to point to your new grammar files.

 

BTW: There is a Python module for speech recognition that uses CMU Sphinx - might be easier for you to use...

 

--- (Edited on 10/4/2016 9:33 am [GMT-0400] by kmaclean) ---

Re: Getting started for dummies
User: andyinyakima
Date: 10/18/2016 12:54 pm
Views: 1207
Rating: 1

I found that using a program called "pavucontrol" allows you to set up your mic levels.

I put the julius sample that I downloaded from VoxForge (thanks VoxForge) on my desktop which I use a handheld mike.

You do have to observe and study the program and the Sample.jconfig file.

I made some changes to my Sample.jconfig to suit me.


# VoxForge configurations:
-dfa grammar/sample.dfa
-v grammar/sample.dict
-h acoustic_model_files/hmmdefs
-hlist acoustic_model_files/tiedlist
-spmodel "sp"        # HMM model name
-multipath
-lv 4000
#-quiet
-input mic
-gprune safe
-iwcd1 max
-iwsppenalty -10.0    # transition penalty for the appended sp models was -70.0
-smpFreq 16000        # sampling rate (Hz)
-iwsp             # append a skippable sp model at all word ends
-penalty1 5.0       
-penalty2 20.0   
-b2 200                 # beam width on 2nd pass (#words)
-sb 200.0        # score beam envelope threshold
-n 1
#-nolog

I started studying the sample.dict file in the /grammar folder, I found that I could add more names. I found out what I did to sample.dict also had to be edited in sample.voca. The VoxForgeDict.txt is where I got my added names.

I've tried to setup PocketSphinx many times and am still trying, but Julius seems to be treating me good. I have tried Kaldi and it is OK but heavy on resources.

andyinyakima

 

--- (Edited on 10/18/2016 12:54 pm [GMT-0500] by andyinyakima) ---

PreviousNext