Got posted on wrong forum by mistake... Wanted to put it in
Speech Recognition forum...
Hi,
I'm looking for an offline tool that can generate Sphinx compatible
english disctionary from a list of words provided (Similar to what
lmtools does). I only want dictionary generation and not the
language model. Which is the suitable tool and can it also support
handdict option for overriding pronunciations as in lmtools?
Li
Last edit: Li3 2012-12-07
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I also suggest you to compile the sphinxtrain from subversion trunk. It has tools to perform your task too - g2p_train and g2p_eval. You just need to compile Sphinxtrain with g2p support.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm interested in this functionality on embedded platform (dictionary to be
used with pocketsphinx). Is logios a suitable choice? If not what should be
the other preferred option. Also could you please provide the most up to date
download links.
Li
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Logios has no sources, you will not be able to compile it for other architecture.
the other preferred option. Also could you please provide the most up to date
download links.
For minimum memory usage and resources letter to sound rules from flite should be the best choice. They are pretty hard to adopt though. Their usage could be less than 100kb.
Current sphinxtrain is more accurate but it requires quite a lot of memory (50MB), so it might not be suitable for you.
Also could you please provide the most up to date download links.
Logios is in the Ravenclaw-Olympus repository and is open-source,
except for one piece which is the letter-to-sound (aka g2p) component.
This is because it is a (very old) piece of MIT software that has a
license attached to it. This is an issue that we haven't resolved as
yet but we are in the process of developing a new version. One option
is to use the g2p included in flite or roll your own using the tools
in festvox. Note that the sphinx trainer includes g2p as well (which
is the same one).
Got posted on wrong forum by mistake... Wanted to put it in
Speech Recognition forum...
Hi,
I'm looking for an offline tool that can generate Sphinx compatible
english disctionary from a list of words provided (Similar to what
lmtools does). I only want dictionary generation and not the
language model. Which is the suitable tool and can it also support
handdict option for overriding pronunciations as in lmtools?
Li
Last edit: Li3 2012-12-07
Hello
Please see the tutorial
http://cmusphinx.sourceforge.net/wiki/tutorialdict
There are many tools to accomplish your task:
Sequitur-g2p, logios
I also suggest you to compile the sphinxtrain from subversion trunk. It has tools to perform your task too - g2p_train and g2p_eval. You just need to compile Sphinxtrain with g2p support.
'll look into the tools you mentioned. Does cmucllmtk also provide this functionality ?
Li
No
Thank you.
I'm interested in this functionality on embedded platform (dictionary to be
used with pocketsphinx). Is logios a suitable choice? If not what should be
the other preferred option. Also could you please provide the most up to date
download links.
Li
Logios has no sources, you will not be able to compile it for other architecture.
For minimum memory usage and resources letter to sound rules from flite should be the best choice. They are pretty hard to adopt though. Their usage could be less than 100kb.
Current sphinxtrain is more accurate but it requires quite a lot of memory (50MB), so it might not be suitable for you.
http://google.com
Last edit: Nickolay V. Shmyrev 2012-12-12
Sources don't exist or not available in public domain ?
Logios is in the Ravenclaw-Olympus repository and is open-source,
except for one piece which is the letter-to-sound (aka g2p) component.
This is because it is a (very old) piece of MIT software that has a
license attached to it. This is an issue that we haven't resolved as
yet but we are in the process of developing a new version. One option
is to use the g2p included in flite or roll your own using the tools
in festvox. Note that the sphinx trainer includes g2p as well (which
is the same one).
Alex
On Wed, Dec 12, 2012 at 4:27 AM, Li3 namhtb@users.sf.net wrote: