I have adapted my acoustic models with new datas and using lm and dic file. Everythings is fine. However, my goal is make adaptation process automatic for my users. For example, if my acoustic/language model is not enough to get sufficient accuracy for my users, they can train by themself using a bat file which is created by me.
Beside this, In my application, user can add some rule on grammar file automatically. However the words which are given by user to grammar file, should be inserted in the dic and language model file. If I don't update the dic, lm file, at that time when I tried to adapt my model, It gives me the "Unable to lookup word 'blabla' in the dictionary" error. So I need to implement some new words automatically in user's language model.
However, language model tools that are recommended in the documentation, require some installation. That means, my every users need to install this tool to be able to add new words in the language model. So this is not practicable way for me. I can not tell them install these tools.
So I have tried to look for any language model tool which doesn't need any installation. But I haven't found yet.
Anyway, I tought that, maybe I don't need to update the language model, and I can adapt the acoustic model using grammar file. For now, It would work for me.
Can I adapt the acoustic model using grammar file ?
What do you suggest me for my case ?
Last edit: kk_huk 2016-06-29
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
You still require your users to install pocketsphinx, so some installation process will be required anyway, you just need to include additional software there.
Beside that there are many language model training tools which you can include into your software as a script. You can check quick_lm.pl script for example. You can also find similar tools for python and other scripting languages. There are distributables toolkits written in Java. So you have plenty of options, you just need to decide what is the programming language you need.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi guys,
I have adapted my acoustic models with new datas and using lm and dic file. Everythings is fine. However, my goal is make adaptation process automatic for my users. For example, if my acoustic/language model is not enough to get sufficient accuracy for my users, they can train by themself using a bat file which is created by me.
Beside this, In my application, user can add some rule on grammar file automatically. However the words which are given by user to grammar file, should be inserted in the dic and language model file. If I don't update the dic, lm file, at that time when I tried to adapt my model, It gives me the "Unable to lookup word 'blabla' in the dictionary" error. So I need to implement some new words automatically in user's language model.
However, language model tools that are recommended in the documentation, require some installation. That means, my every users need to install this tool to be able to add new words in the language model. So this is not practicable way for me. I can not tell them install these tools.
So I have tried to look for any language model tool which doesn't need any installation. But I haven't found yet.
Anyway, I tought that, maybe I don't need to update the language model, and I can adapt the acoustic model using grammar file. For now, It would work for me.
Can I adapt the acoustic model using grammar file ?
What do you suggest me for my case ?
Last edit: kk_huk 2016-06-29
You still require your users to install pocketsphinx, so some installation process will be required anyway, you just need to include additional software there.
Beside that there are many language model training tools which you can include into your software as a script. You can check
quick_lm.plscript for example. You can also find similar tools for python and other scripting languages. There are distributables toolkits written in Java. So you have plenty of options, you just need to decide what is the programming language you need.Thanks a lot Nickolay,
I have checked the quick_lm.pl script as you said. Happily, It works for me.