Hi every one. If have a serious problem.
I have setup a sphinxtrain system on my winxp system and succesfully train my own acoustic model. Everything works fine and the decoding works to. Now i have the posseblity to do some modeling on a fast linux system in universety. So setup sphinxtrain and trained the an4 model for a test. Still everything is fine and fast now.
Now i put an my data on the linux system and my problems start. My dic file has about 2000 words and and the verify of my data is says everythings ok but i step 3 when generating the untied mdef i get about 2500 warnings that my words are mapped to silence because word not found in dictionary.
And because of that there is no initialization for the next step and everythings crashes. Does anyone has an idea why?
My dic looks just as the an4.dic and has all rights.
hope someone cloud help me.I can send you my data if you need it.
nesayus
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2006-07-12
I suspect that your problem is the end-of-line character(s) in your dictionary file (and all other text files). Your first training was done on WinXP, and now you are using a Linux system. I have not looked at the SphinxTrain code for over a year, but as I recall, the text-file reading code is not set up to handle all the possible end-of-line characters that are used in various operating systems. This has been a problem for a number of people over the past few years.
I hope that helps.
cheers,
jerry wolf
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I've delted the CR at the end of the lines. I've done that before but it did not work. So I took an other program to copy the data (ssh secure shell instead of winSCP 3). Now it trains :)
Thx a lot!
nesayus
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi every one. If have a serious problem.
I have setup a sphinxtrain system on my winxp system and succesfully train my own acoustic model. Everything works fine and the decoding works to. Now i have the posseblity to do some modeling on a fast linux system in universety. So setup sphinxtrain and trained the an4 model for a test. Still everything is fine and fast now.
Now i put an my data on the linux system and my problems start. My dic file has about 2000 words and and the verify of my data is says everythings ok but i step 3 when generating the untied mdef i get about 2500 warnings that my words are mapped to silence because word not found in dictionary.
And because of that there is no initialization for the next step and everythings crashes. Does anyone has an idea why?
My dic looks just as the an4.dic and has all rights.
hope someone cloud help me.I can send you my data if you need it.
nesayus
I suspect that your problem is the end-of-line character(s) in your dictionary file (and all other text files). Your first training was done on WinXP, and now you are using a Linux system. I have not looked at the SphinxTrain code for over a year, but as I recall, the text-file reading code is not set up to handle all the possible end-of-line characters that are used in various operating systems. This has been a problem for a number of people over the past few years.
I hope that helps.
cheers,
jerry wolf
I've delted the CR at the end of the lines. I've done that before but it did not work. So I took an other program to copy the data (ssh secure shell instead of winSCP 3). Now it trains :)
Thx a lot!
nesayus