Can I create phoneme and intregrate it in sphinx ?
If it possible, how ?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-24
Is there a software allowing of generate files vec var map and phone ?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-24
Your questions are rather vague and therefore hard to answer. With respect to your second question, "files vec var map and phone" refers to some of the files in a Sphinx2 acoustic model, so you appear to be asking how to create an acoustic model.
Acoustic models are created by processing a large dataset of transcribed utterances, together with a definition of a phone set and a dictionary of all the words in the utterances, giving their pronunciations in terms of the defined phones. In additon to Sphinx2 and Sphinx3, CMU has made available a software suite and a set of perl scripts called SphinxTrain, which can be used to produce an acoustic model for Sphinx2 or 3. See http://cmusphinx.sourceforge.net/webpage/html/cmusphinx.html and http://www.speech.cs.cmu.edu/SphinxTrain/index.html .
There is documentation available at those URLs. Unfortunately, the model training process is quite complex, and the documentation is incomplete and difficult to understand, and so the training of an acoustic model from a dataset of speech is not easy to do unless you have a background in HMM speech recognition.
The Sphinx2 and Sphinx3 open source distributions each contain an acoustic model for American English fluent adult male and female speech, but if your interest is in some other language, that would be no good to you. Good luck!
cheers,
jerry wolf
soliloquy learning, inc.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-25
Thanks
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-26
Sphinx Train remake all but it isn't what I want.
I'm use Sphinx II in allphone mode and I create a program which realize a "word spoting" recognition for french language. But several chmm file miss.
I'm found them but how insert them in sphinx files without remake all acoustic model
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-26
You have said that several of the .chmm files in the acoustic model are missing, and that you have found them (or copies of them), but you don't know where to put them. Is that correct?
Here is a listing of one of my Sphinx2 acoustic models, which consists of about 55 files in one directory. It contains (1) phone, (2) map, (3) sendump, (4) eight codebook files, and (5) about 50 .chmm files:
There are five Sphinx2 arguments that refer to these five types of model files (-phnfn, -mapfn, -sendmpfn, -cbdir, and -hmmdir). Specifically, -hmmdir specifies the directory that holds all the .chmm files. Therefore, put the new .chmm files in that directory along with the other .chmm files.
Is that the information you were asking for?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-26
But the new chmm files aren't reference in the others (map,phone,vec and var).
I'm not sure that they update themself !?!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-26
(I think I made a mistake in my message above. The Sphinx2 argument -hmmdirlist gives the directory containing the .chmm files. The -hmmdir argument has another use, which I don't understand...)
Sorry, I still don't understand your reply and your problem. Sphinx2 reads all the .chmm files (one file for each CI phone) near the end of the function kb().
The .chmm files are part of the Sphinx2 acoustic model, and Sphinx2 does not update or modify them.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-26
The map file, for example, no contain your new chmm file.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-26
The phone file in the Sphinx2 acoustic model specifies all the filler and context-independent (CI) phones and all the triphones in the model. The first ~50 lines specify all the filler and CI phones, such as +BREATH+, AA, AE, etc. For each such phone, Sphinx then reads in a .chmm file from the -hmmdirlist directory. That's how it knows which .chmm files to read.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-27
If I have only the chmm files, how generate others files of this directory?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-27
The .chmm files contain only the topology and the transition probabilities of the HMMs for the triphones and CI phones. Given only these files it is not possible to generate the other files of the acoustic model. For that, you would need to use SphinxTrain to train a complete acoustic model from a dataset of transcribed speech, as I mentioned in a message earlier this week.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2004-05-27
Thanks
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I did a word spotter in French by using the provided french models and it works great.
By the way it works in english, too.
What I do is dumping the lattice file. It contains all the hypotheses frame after frame, with an absolute score. Based on this score I decide wether the word has been pronounced in the sentence or not.
dewavrin at yahoo.com
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Can I create phoneme and intregrate it in sphinx ?
If it possible, how ?
Is there a software allowing of generate files vec var map and phone ?
Your questions are rather vague and therefore hard to answer. With respect to your second question, "files vec var map and phone" refers to some of the files in a Sphinx2 acoustic model, so you appear to be asking how to create an acoustic model.
Acoustic models are created by processing a large dataset of transcribed utterances, together with a definition of a phone set and a dictionary of all the words in the utterances, giving their pronunciations in terms of the defined phones. In additon to Sphinx2 and Sphinx3, CMU has made available a software suite and a set of perl scripts called SphinxTrain, which can be used to produce an acoustic model for Sphinx2 or 3. See http://cmusphinx.sourceforge.net/webpage/html/cmusphinx.html and http://www.speech.cs.cmu.edu/SphinxTrain/index.html .
There is documentation available at those URLs. Unfortunately, the model training process is quite complex, and the documentation is incomplete and difficult to understand, and so the training of an acoustic model from a dataset of speech is not easy to do unless you have a background in HMM speech recognition.
The Sphinx2 and Sphinx3 open source distributions each contain an acoustic model for American English fluent adult male and female speech, but if your interest is in some other language, that would be no good to you. Good luck!
cheers,
jerry wolf
soliloquy learning, inc.
Thanks
Sphinx Train remake all but it isn't what I want.
I'm use Sphinx II in allphone mode and I create a program which realize a "word spoting" recognition for french language. But several chmm file miss.
I'm found them but how insert them in sphinx files without remake all acoustic model
You have said that several of the .chmm files in the acoustic model are missing, and that you have found them (or copies of them), but you don't know where to put them. Is that correct?
Here is a listing of one of my Sphinx2 acoustic models, which consists of about 55 files in one directory. It contains (1) phone, (2) map, (3) sendump, (4) eight codebook files, and (5) about 50 .chmm files:
AA.chmm CH.chmm G.chmm OW.chmm T.chmm
AE.chmm d2cep.256.var HH.chmm OY.chmm TH.chmm
AH.chmm d2cep.256.vec IH.chmm p3cep.256.var UH.chmm
AO.chmm D.chmm IX.chmm p3cep.256.vec UW.chmm
AW.chmm DH.chmm IY.chmm P.chmm V.chmm
AX.chmm DX.chmm JH.chmm phone W.chmm
AXR.chmm EH.chmm K.chmm R.chmm xcep.256.var
AY.chmm ER.chmm L.chmm S.chmm xcep.256.vec
B.chmm EY.chmm map sendump Y.chmm
+BREATH+.chmm F.chmm M.chmm SH.chmm Z.chmm
cep.256.var +FP+.chmm N.chmm SIL.chmm ZH.chmm
cep.256.vec +FRG+.chmm NG.chmm +SN+.chmm
There are five Sphinx2 arguments that refer to these five types of model files (-phnfn, -mapfn, -sendmpfn, -cbdir, and -hmmdir). Specifically, -hmmdir specifies the directory that holds all the .chmm files. Therefore, put the new .chmm files in that directory along with the other .chmm files.
Is that the information you were asking for?
But the new chmm files aren't reference in the others (map,phone,vec and var).
I'm not sure that they update themself !?!
(I think I made a mistake in my message above. The Sphinx2 argument -hmmdirlist gives the directory containing the .chmm files. The -hmmdir argument has another use, which I don't understand...)
Sorry, I still don't understand your reply and your problem. Sphinx2 reads all the .chmm files (one file for each CI phone) near the end of the function kb().
The .chmm files are part of the Sphinx2 acoustic model, and Sphinx2 does not update or modify them.
The map file, for example, no contain your new chmm file.
The phone file in the Sphinx2 acoustic model specifies all the filler and context-independent (CI) phones and all the triphones in the model. The first ~50 lines specify all the filler and CI phones, such as +BREATH+, AA, AE, etc. For each such phone, Sphinx then reads in a .chmm file from the -hmmdirlist directory. That's how it knows which .chmm files to read.
If I have only the chmm files, how generate others files of this directory?
The .chmm files contain only the topology and the transition probabilities of the HMMs for the triphones and CI phones. Given only these files it is not possible to generate the other files of the acoustic model. For that, you would need to use SphinxTrain to train a complete acoustic model from a dataset of transcribed speech, as I mentioned in a message earlier this week.
Thanks
Hi,
I did a word spotter in French by using the provided french models and it works great.
By the way it works in english, too.
What I do is dumping the lattice file. It contains all the hypotheses frame after frame, with an absolute score. Based on this score I decide wether the word has been pronounced in the sentence or not.
dewavrin at yahoo.com