Simon: Problems while compiling the speech model
Page 1 of 1 (13 posts)
Tags:
None
Registered Member 
|
Hello everyone,
Simon has problems to compile the speech model after I upgraded to 13,04. I realized that it takes simon a lot of time to compile a speech model after a training and it doesn´t come to an ending. The training was 54 words long, so I just thought that this is normal. I started Simond (in my case KSimond) in the bash and realized that that Simon hat the same error all of the time:
This message goes on forever (until you kill the process). Here is some information about my system: Lubuntu 13.04 64-Bit Kernel 3.8.0-21-generic Simon 0.4.0 sphinxtrain 1.0.8 spinxbase 0.8 pocketsphinx 0.8 I reinstalled Simon, the Sphinx components and julius but it didn´t solve the problem. I also tried to install HTK instead of Sphinx but I get some errors while compiling. Greetings |
|
Moderator
|
Without more information, this is hard to diagnose.
Can you possibly tar up /tmp/kde-clemens/simond/default/compile/sphinx/ after a failed compilation and upload it somewhere? Thanks. Best regards, Peter |
|
Registered Member
|
Hello,
default-82ce3268-dbc8-43f2-a648-63f1b193e1b2-.tar . This is the link to the file you asked. To be honest, I don´t understand much of the content of the folder. Greetings |
|
Moderator
|
Thanks for the data, this was extremely helpful.
You uncovered some crude dictionary handling in the sphinx adaption engine that caused it to mess up when confronted with the German "ß" (because toLower(toUpper("ß")) != "ß"). I was able to reproduce your errors with the German "Tastatur" scenario and fixed them so that everything now builds and works fine for me. If you want, you can build Simon from git and it should "just work" (it would be great if you would try that and confirm). However, the changes will also land in Simon 0.4.1 to be released next Monday. In any case, thank you again, this report really helped a lot as I probably wouldn't have noticed that before 0.4.1. Best regards, Peter |
|
Registered Member
|
Hello,
thank you very much of your effort. I the code from Sourceforge with git but I get an error message while compiling simon:
I used the following adress: "git clone git://speech2text.git.sourceforge.net/g ... peech2text speech2text" I cloned the code two times, it's unlikely that I have a corrupted file. The error log might be usefull, I just wait till Monday. Thank you a lot anyway Greetings, Clemens |
|
Moderator
|
The current Simon version is to be found on git://anongit.kde.org/simon.git - not on Sourceforge.
However the error clearly states that you are missing zlib (or its development files). From this, I take it that you were using the binary packages? Those will probably not be released on Monday; I have no influence on when Ubuntu will ship updated binaries, I can only tell you when the new version (in source form) will be available. Best regards, Peter |
Administrator
|
You need to install the appropriate "ZLib" development package using your package manager. On openSUSE at least, this is called "zlib-devel", but may differ on other distributions.
KDE Sysadmin
[img]content/bcooksley_sig.png[/img] |
|
Registered Member
|
Hello everyone,
thank you all very much for your effort. I cloned he program with git from "git://anongit.kde.org/simon.git". There were no problems while compiling the programm but I got problems again while compiling the speech model. I get the same output in the teriminal for three times. after the first and the second try is a message "try again: true". after the third time "try again: false". There is also a new error window in simon popping up. I gues this is a new improvement in the new version. It says the following text:
Good night, Clemens |
|
Moderator
|
Nope, I can't reproduce that at all here.
Please send me another archive of your current /tmp/kde-clemens/simond/default/compile/sphinx/ Thanks. Best regards, Peter |
|
Registered Member
|
Hello,
this time it´s only 2 Mb big. Hope you find what you search: default-dfccb977-caea-4e65-ad55-7ad7555212b8-.tar Greetings, Clemens |
|
Moderator
|
Thanks. I did.
Please try again. Best regards, Peter |
|
Registered Member
|
Hello,
thank you a lot. It works now. The strange thing is that it only works in the third try. It takes a bit long. Greetings, Clemens P.S.: The simon Webistes show a MySQL error on their pages. Don´t know if you are the admin of these pages |
|
Moderator
|
Perfect.
That it works only on the third try is actually expected. Here's the deal: The keyboard scenario is a bit tricky because it's rather large (lots of words) and we want it to work asap (little training data). So we include alternate pronunciations in the scenario (e.g.: "Sieben" may be pronounced either like it is written or as "Siebn"). While compiling, the forced alignment will pick the better fitting of those pronunciations and train the model based on that. Because the model is too small to get good phoneme coverage the not-picked transcriptions will cause phonemes that are in the model but have not been trained (e.g. the nasal "n" from the word "Siebn" may not be be observed in your training data). This causes the training to fail. But, Simon picks up on the error, and fixes it by removing the missing phoneme (and with it the alternate transcription that wasn't picked anyways). It then starts the compilation again. In your case, you have two of such "errors" in succession, which is why it takes a little while. The removal of the alternate transcriptions is temporary, btw. So if you train the alternate pronunciation sometime in the future, the full set of possible transcriptions will be considered. Best regards, Peter Ps.: simon-listens.org is the homepage of the organization Simon Listens. I don't have access there as I am no longer a member. If you're looking for updates on Simon, you may want have a look at the Simon website: http://simon.kde.org |
Page 1 of 1 (13 posts)
Bookmarks
Who is online
Registered users: Bing [Bot], claydoh, Google [Bot], rblackwell
