Hello

I finally have julius working, and julia, and htk.

I did the simple 'call/dial' 'steve/young' and numbers and it came out pretty good.  I've changed to a word loop system to allow a more dictation approach over IVR but with a limited dictonary of 140 words directly related to the end use case.

The end use case being a jasper system running on PI connected to the NMEA2000 can bus network on my boat.  The idea being that i can vebally request information from the boat and have it give it to me, a la StarTrek: TNG :)

So i have this 140 some odd words all revolving around these commands i'd like to use but the sample dictonary set up doesn't really cover it.  Its catching 1 in 20 or so of those words.  ergo, i need to know how to set up more training files.  I have the 40 sample scripts but they don't seem to cover all the phonems i need.  I just don't know how to either add to or modify those (i would presume add to).

Logically, i'm guessing, i should just make vocal samples based on each of the commands i might say to the system.  There's a good 30 of those alone.  If i say them forward and backwards thats 60 more sample wavs i can work with.  100 total.

Thing is... how do?  


Also, i'd like to try to adopt a grammar system that follows:
Activator: Command: word loop
Computer, turn ...  (left/on lights/ etc) to help try my accuracy of responses.  I'm not sure if it matters with the jasper engine behind it though, so if you got any experience with that i'm all ears.

Thanks for all the help so far guys.  I got it hearing 'computer' pretty reliably :)  Onward and upward from here.

--- (Edited on 12/22/2020 10:51 pm [GMT-0600] by maglinvinn) ---