Menu

Training Sphinx on runtime

Help
tthung
2012-03-10
2012-09-22
  • tthung

    tthung - 2012-03-10

    I'm totally a newbie here so maybe my question is too dummy. My question is:

    I want to build an application base on Sphinx that can allow user to train
    this application with their voice or their commands. The application on next
    time will remember their voice or command and do whatever the user defined
    before. Any one gives me the solution or the other way instead of using Sphinx
    ?

     
  • Nickolay V. Shmyrev

    Hello tonthanhhung

    Welcome to our project

    I want to build an application base on Sphinx that can allow user to train
    this application with their voice or their commands. The application on next
    time will remember their voice or command and do whatever the user defined
    before.

    It's easy to build such an application. Actually there are quite some examples
    of it already which you can find on our website:

    http://cmusphinx.sourceforge.net/wiki/sphinxinaction

    You can start learning about CMUSphinx by reading a tutorial

    http://cmusphinx.sourceforge.net/wiki/tutorial

     
  • tthung

    tthung - 2012-03-11

    Some of the examples you gave on the link are built on English Acoustic model.
    The programmer just need to apply that model to their project.
    I want to create the application that can be trained in run-time, whatever
    language the users use. It remember what be learnt.
    For example, the user say a word "ni hao" (Chinese word means "Hello") for ten
    times or over, the application of couse will doesn't understand the meaning,
    but it will recognize the "ni hao" sound for the next time the speaker talk.
    I know a little about Sphinx, it uses speech feature to built HMM, and use
    some HMM's algorithms to do recognition,
    Can Sphinx use "Vector quantization" to do recognition ?

     
  • Nickolay V. Shmyrev

    Some of the examples you gave on the link are built on English Acoustic
    model. The programmer just need to apply that model to their project.

    Some of them are, some not. Check other parts of the tutorial

    I want to create the application that can be trained in run-time, whatever
    language the users use. It remember what be learnt. For example, the user say
    a word "ni hao" (Chinese word means "Hello") for ten times or over, the
    application of couse will doesn't understand the meaning, but it will
    recognize the "ni hao" sound for the next time the speaker talk.

    Nothing stops you from doing that

    I know a little about Sphinx, it uses speech feature to built HMM, and use
    some HMM's algorithms to do recognition,

    You have a great chance to learn more

    Can Sphinx use "Vector quantization" to do recognition ?

    Vector quantization is used in CMUSphinx in several places to optimize the
    computation. In fast GMM computation, in semi-continuous mixture weight
    compression and in some other places. Which particular usage do you have in
    mind?

     
  • tthung

    tthung - 2012-03-12

    Thank you so much for helping me, nshmyrev. I will try more... can you give me
    some material to help me learn some basic knowledge to complete my project
    (free source plz).

     

Log in to post a comment.