As I am interested in phoneme recognition only, I found Allphone functions quite interesting. But, mfc files are needed to launch allphone demo. How can I get thoses files; I found a link on a wave2mfc converter at CMU but it was broken: where can I get such a program ?
Lionel Deglise
(French student)
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Are you interested in alignment or in allphone decode? What we have for the phone decode isn't very accurate -- the recognizer does much better with more constraints.
Are you looking to find the timing of a transcription you have, or really wanting to run it as allphone decoding?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2000-02-08
We really want to do allphone decoding. Our goal is to get a continous phone transcription of a continous speech. Accurancy can be average.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
OK, accuracy is not that great, but i've added an all-phone decode example to the CVS tree. Go ahead and check out the tree, run ./autogen.sh, and then ./configure, make, make install, and run sphinx2-phone. sphinx2-phone is a shell script that calls sphinx2-allphone with some paramters -- including -datadir, which is the directory the data file lives in.
it will ask for a file name, without an extension (this is taken care of with -adcin TRUE -adcext 16k for the example file). By default it's looking in /usr/share/sphinx2/lm/turtle, and there's a file in there called goforward.16k. Tell sphinx2-phone that you want to decode
goforward
and it will decode that file. Note the accuracy is not that great. Nonetheless, it is an example of how to do allphone decode. You can use the -ctlfile option to make a control file of filenames in the data dir (without extension) -- this will be in the forthcoming docs more clearly.
kevin
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
As I am interested in phoneme recognition only, I found Allphone functions quite interesting. But, mfc files are needed to launch allphone demo. How can I get thoses files; I found a link on a wave2mfc converter at CMU but it was broken: where can I get such a program ?
Lionel Deglise
(French student)
Are you interested in alignment or in allphone decode? What we have for the phone decode isn't very accurate -- the recognizer does much better with more constraints.
Are you looking to find the timing of a transcription you have, or really wanting to run it as allphone decoding?
We really want to do allphone decoding. Our goal is to get a continous phone transcription of a continous speech. Accurancy can be average.
OK, accuracy is not that great, but i've added an all-phone decode example to the CVS tree. Go ahead and check out the tree, run ./autogen.sh, and then ./configure, make, make install, and run sphinx2-phone. sphinx2-phone is a shell script that calls sphinx2-allphone with some paramters -- including -datadir, which is the directory the data file lives in.
it will ask for a file name, without an extension (this is taken care of with -adcin TRUE -adcext 16k for the example file). By default it's looking in /usr/share/sphinx2/lm/turtle, and there's a file in there called goforward.16k. Tell sphinx2-phone that you want to decode
goforward
and it will decode that file. Note the accuracy is not that great. Nonetheless, it is an example of how to do allphone decode. You can use the -ctlfile option to make a control file of filenames in the data dir (without extension) -- this will be in the forthcoming docs more clearly.
kevin