For a lot of time i have been trying to create the decision trees, recently i succeeded in doing so but some of the phonemes for which there were no triphones, no decison trees could be generated. to trace the cause i dived into the code, and in main.c (int bldtree directory in src/programs) there is i think an error. On seeing that there are no triphones the program returns a success, without initializing the n-model variable, this causes a garbage value to be passed onto ckd_calloc and a fatal error occurs.
I may be wrong but i wanted to bring my find into your consideration.
Secondly would you be kind enough to tell me that how n_model is initialized in the init function (in terms of theoretical concepts..like all the distributions and calculations involved..)
Regards
BILAL AHMED
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I also reply you in another thread that your problem is not really caused by this bug. Please kindly read my replies and suggestions in another thread. I will strongly recommend you not to use audio novel as your first task.
It will just give you insufficient data to train a partcular phoneme. At the end, you training will still not work out because of that.
From now on please allow me to just take care of your bug report coz I really can't support a processo of training which I don't know how to support.
-Arthur
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thnkyou for your consideration even after the fact that i am adamant about creating my own corpus. One of the basic driving forces behind all this efforts is the fact that people here in my country do not have access to big corporas like LDC (TIDIGITS AND TIWORDS), if i succeed in making one i could help others and as my current task is trivial, i could concecntrate on building a corpus.
Regards
BILAL AHMED
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Bilal,
As I have urged you to read my mail, I hope you can kindly read it another time.
There exist free databases, one of them is called AN4 and was distributed freely by CMU. You can **download** it at here, http://www.speech.cs.cmu.edu/sphinx/database
So unlikely TIDIGITS and TIWORDS, it is **free**. Again, I hope that can help you.
Arthur
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
For a lot of time i have been trying to create the decision trees, recently i succeeded in doing so but some of the phonemes for which there were no triphones, no decison trees could be generated. to trace the cause i dived into the code, and in main.c (int bldtree directory in src/programs) there is i think an error. On seeing that there are no triphones the program returns a success, without initializing the n-model variable, this causes a garbage value to be passed onto ckd_calloc and a fatal error occurs.
I may be wrong but i wanted to bring my find into your consideration.
Secondly would you be kind enough to tell me that how n_model is initialized in the init function (in terms of theoretical concepts..like all the distributions and calculations involved..)
Regards
BILAL AHMED
Hmm. I will also call it a bug. Please file a Bug report. You can file a bug report in the "Bugs" tab of this web page.
-Arthur
I also reply you in another thread that your problem is not really caused by this bug. Please kindly read my replies and suggestions in another thread. I will strongly recommend you not to use audio novel as your first task.
It will just give you insufficient data to train a partcular phoneme. At the end, you training will still not work out because of that.
From now on please allow me to just take care of your bug report coz I really can't support a processo of training which I don't know how to support.
-Arthur
Thnkyou for your consideration even after the fact that i am adamant about creating my own corpus. One of the basic driving forces behind all this efforts is the fact that people here in my country do not have access to big corporas like LDC (TIDIGITS AND TIWORDS), if i succeed in making one i could help others and as my current task is trivial, i could concecntrate on building a corpus.
Regards
BILAL AHMED
Bilal,
As I have urged you to read my mail, I hope you can kindly read it another time.
There exist free databases, one of them is called AN4 and was distributed freely by CMU. You can **download** it at here,
http://www.speech.cs.cmu.edu/sphinx/database
So unlikely TIDIGITS and TIWORDS, it is **free**. Again, I hope that can help you.
Arthur