I am currently using the Helloworld application for testing. I understand that the .gram file is where I need to add new words for it to recognize.
My question is: Instead of manually adding words to the .gram file, how do i make Sphinx4 recognize any word i speak (which is present in the dictionary)?
There is a dictionary file - cmu0.5.dict - I just want to configure the helloworld program to recognize any word in that dictionary and not the hello.gram file.
Thanks
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
It's almost impossible. There are hundred thousands words in cmudict, you will not build a recognizer for such a big vocabulary.
Define another task for yourself. Something different than "recognize every word I say". Be more realistic, what kind of text do you want to decode, what is the size of the vocabulary? It's possible to decode texts of around 20000 words but not more.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
OK...Assume that I keep only 10000 words in the cmudict file, what configuration changes do i have to make to the helloworld program for it to recognize from the dict file and not from the hello.gram file (where i need to manually enter words)?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I dont have any text of my own to train. The example that you gave (http://www.mediafire.com/download.php?5uxxffxsjop) has medical terms....i just want to use simple spoken english words.
I just want to use the default words that comes in the cmu0.5dict file.
Can you let me know what changes I have to make in the config xml in the helloworld application to use the dict file (i will cut down the words to 10000 instead of existing 0.1M records), instead of grammar(hello.gram) file?
Basically, I want to have a continuous speech processing program wherein if I speak 'N' number of random words at a time, all should be recognized. NOTE that all the N words that I speak ARE PRESENT in the truncated dictionary cmu0.5dict.
Thanks.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi All,
Sphinx4 + Eclipse
Windows XP SP2
I am currently using the Helloworld application for testing. I understand that the .gram file is where I need to add new words for it to recognize.
My question is: Instead of manually adding words to the .gram file, how do i make Sphinx4 recognize any word i speak (which is present in the dictionary)?
There is a dictionary file - cmu0.5.dict - I just want to configure the helloworld program to recognize any word in that dictionary and not the hello.gram file.
Thanks
Alright, thanks.
Is there a tutorial available as to how I can load the model into my XML file?
It's almost impossible. There are hundred thousands words in cmudict, you will not build a recognizer for such a big vocabulary.
Define another task for yourself. Something different than "recognize every word I say". Be more realistic, what kind of text do you want to decode, what is the size of the vocabulary? It's possible to decode texts of around 20000 words but not more.
OK...Assume that I keep only 10000 words in the cmudict file, what configuration changes do i have to make to the helloworld program for it to recognize from the dict file and not from the hello.gram file (where i need to manually enter words)?
You have to train language model for your texts then and use trigram linguist instead of jsgf one. See HelloNGram and http://www.mediafire.com/download.php?5uxxffxsjop for example.
Hi,
I dont have any text of my own to train. The example that you gave (http://www.mediafire.com/download.php?5uxxffxsjop) has medical terms....i just want to use simple spoken english words.
I just want to use the default words that comes in the cmu0.5dict file.
Can you let me know what changes I have to make in the config xml in the helloworld application to use the dict file (i will cut down the words to 10000 instead of existing 0.1M records), instead of grammar(hello.gram) file?
Basically, I want to have a continuous speech processing program wherein if I speak 'N' number of random words at a time, all should be recognized. NOTE that all the N words that I speak ARE PRESENT in the truncated dictionary cmu0.5dict.
Thanks.
Well, you can load 5k model here for example. I'm not sure it will work with sphinx4:
http://www.inference.phy.cam.ac.uk/kv227/lm_giga/