I'm learning Sphinx4 at the moment. I'm trying to cut off the part of the
project that recognizes separate phonemes. I do not need to make words from
them. Another words, I want to count appearences of phonems in the input
stream. I learned how frontend package proccess the data, but I have no idea
how searcher package manage that data. Is there any
presentetation/tutorial/book for that algorithm? share some info on that if
you be so kind :) sory for my last post. I have to spend a little more time
reading /wiki, to have all the answers.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
To be more clear. My target is to detect some phonems (not all) and to change
their spectrum. I'm thinking of make a tree in front-end after "fft"
DataProcessor. One branch is for recognition, another changes spectrum (or
not) depending on results of recognition. After that, the second branch is
under inverse FFT and than data frames are collected back in the stream. (Idea
is simple, but I'm not I told it in simple way) Is it possible? Where I can
learn the structure of your recognizer to cut off the part I need?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks for fast reply. gonna try that!! Awesome that such project has support,
because it can take monthes to deal with it even for a simpliest task!
Спасибо!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm learning Sphinx4 at the moment. I'm trying to cut off the part of the
project that recognizes separate phonemes. I do not need to make words from
them. Another words, I want to count appearences of phonems in the input
stream. I learned how frontend package proccess the data, but I have no idea
how searcher package manage that data. Is there any
presentetation/tutorial/book for that algorithm? share some info on that if
you be so kind :) sory for my last post. I have to spend a little more time
reading /wiki, to have all the answers.
To be more clear. My target is to detect some phonems (not all) and to change
their spectrum. I'm thinking of make a tree in front-end after "fft"
DataProcessor. One branch is for recognition, another changes spectrum (or
not) depending on results of recognition. After that, the second branch is
under inverse FFT and than data frames are collected back in the stream. (Idea
is simple, but I'm not I told it in simple way) Is it possible? Where I can
learn the structure of your recognizer to cut off the part I need?
You need to create you own Linguist, basically to implement the Linguist
interface. You can read more about linguists at
http://cmusphinx.sourceforge.net/sphinx4/javadoc/edu/cmu/sphinx/linguist/Ling
uist.html
That linguist should return the following search graph as a search space:
http://cmusphinx.sourceforge.net/sphinx4/javadoc/edu/cmu/sphinx/linguist/flat
/CIPhoneLoop.PhoneLoopSearchGraph.html
It's possible. To learn sphinx4 architecture read the main document:
http://cmusphinx.sourceforge.net/sphinx4/doc/Sphinx4Whitepaper.pdf
and you might be also interested in some other documents in the doc folder in
sphinx4 sources.
Thanks for fast reply. gonna try that!! Awesome that such project has support,
because it can take monthes to deal with it! Спасибо!
Thanks for fast reply. gonna try that!! Awesome that such project has support,
because it can take monthes to deal with it even for a simpliest task!
Спасибо!