Dojo Foundation: can anyone help me get started?
Ohmu
I am trying to write an iPhone / iPad application that teaches this
system to a student.
my system extends the traditional 'Do Re Mi'.
I am using 12 consonant syllables and 12 vowel syllables
each musical note consists of a consonant ( to represent the
physical note ) followed by a vowel ( to represent the position
relative to the tonal centre )
so that is 144 possible combinations, like 'bah' 'goo' 'tii'
'lur' ...
you can see it in action on the first picture on my website: www.toneme.org
firstly, a big thank you to Brian and everyone else who has created
this project. I have severe RSI and limited technical ability, and
this represents weeks of solid hammering on the computer. At least! It
is wonderful of you guys to put this up for free. If it were not for
this project, my goal would surely be beyond my reach.
Could anyone who understands the internals of this project offer me
some guidance on how I can accomplish what I want to do?
Basically I want to set up this engine so that it listens out for
these 144 phoneme pairs, and reports back the instant it detects one.
My first question is: training.
* do I have to create training data which would ship with the
application? if so, how?
* Would the user have to go through a training process the first time
they run the app?
second question:
* how much of this code do I need? I am guessing maybe I could gut
out the lower levels. I don't need trigram matching, I don't
think.
any help / guidance that you can give would be hugely appreciated!
I know that I am going to have to read through a lot of sphinx
documentation, but it would help me immeasurably to first
understand the task in hand. That way, when I read, I am not reading
blind and stumbling in the dark, rather I would be starting out with
an understanding and reading to learn the specifics
Sam
sunf...@gmail.com
Brian King
This project looks like a great idea! I'd love to help you hash
out idea's of an approach that might work, a few things that I notice:
1) It appears that this would require very real-time continuous
dictation, something that I'm not sure pocketsphinx is up for. There
is some code to support continuous dictation, but I know this can suck
some CPU time. And while it can do "Continuous" dictation, it will
not have a real-time response. Decoding one word on an iPhone 4 is
rarely under a second. Also, I imagine if the notes were held for a
while, they would probably be goofed a bit.
2) It would need a custom dictionary that should be pretty easy to
generate. You don't have to do any training or anything, just create
a custom language model.
http://cmusphinx.sourceforge.net/wiki/languagemodelhowto
3) You do need all the code. Well you might not, but trimming it out
will be more effort than it's worth
4) You may be able to do this entirely based on pitch with visual
feedback. It would slightly restrict your feature set, but if you're
interested in real time results it should be very feasible. Sure, if
you said Do out of pitch, Do would not light up, but the visual
feedback may be able to provide the training just because the person
is saying one thing and seeing another light up. I really think this
the direction to go in. But your vision in that document was
substantially over my head and I'm not sure of all of the use cases
you want to see.
If you wanted to talk with me about your application usability and how
to make it pitch based, email me off list!
Brian King
> --
> You received this message because you are subscribed to the Google Groups "VocalKit" group.
> To post to this group, send email to voca...@googlegroups.com.
> To unsubscribe from this group, send email to vocalkit+u...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/vocalkit?hl=en.
>
>