Salve a tutti. Sono giorni che leggo il vostro forum e innanzitutto vi voglio fare i complimenti per ciò che fate. Vorrei sapere se qualcuno può aiutarmi. Vorrei far funzionare la demo "hello world" di sphinx4 in linguaggio JAVA, facendogli riconoscere parole italiane. Per tale scopo vorrei addestrare un modello in italiano, anche se ovviamente mi rendo conto che con pochi dati è una cosa molto difficile. Il primo problema è: qualcuno sa come impostare il file .xml o quali impostazioni cambiare per far funzionare la demo con un'altra lingua? Ho letto già molte guide sul sito di sphinx e molti altri suggerimenti su diversi forum, ma non mi è stato di grande aiuto. Ringrazio anticipatamente chi vorrà aiutarmi.

Download latest demo

http://github.com/cmusphinx/sphinx4

Build it with maven, run TranscriberDemo from sphinx4-samples package.

There are no XML files anymore, to configure it with Italian you need to download Italian acoustic model from Voxforge:

AcousticModels.tgz

and configure the model in the source code:

configuration.setAcousticModelPath("file:voxforge_it.cd_cont_1000");

configuration.setDictionaryPath("file:voxforge_it.dict");

configuration.setLanguageModelPath("file:voxforge_it.lm.dmp");

You can learn more about models from CMUSphinx tutorial

http://cmusphinx.sourceforge.net/wiki/tutorial

thanks for your quick reply. I'm trying to recognize speech from a microphone and not a wav file, so the demo I'm using is the hello world. Can I do the same thing for that? Also I need the program to recognize words that are not in the dictionary as unknown, can I activate the OutOfGrammar probability with a foreign language? So I think if I configure the acoustic model with this 3 lines of code:

configuration.setAcousticModelPath("file:voxforge_it.cd_cont_1000");

configuration.setDictionaryPath("file:voxforge_it.dict");

configuration.setLanguageModelPath("file:voxforge_it.lm.dmp");

I must eliminate this,

cm = new ConfigurationManager(SpeechRecognizer.class.getResource("speechRecognizer.config.xml"));

right?

So How can I activate OutOfGrammar probability in this case?

> so the demo I'm using is the hello world

The demo for microphone recognition is called Dialog, we do not have hello world anymore

> I must eliminate this, cm = new ConfigurationManager(SpeechRecognizer.class.getResource("speechRecognizer.config.xml"));

right?

There is no such thing in latest code

thanks for the clarification. I worked with the previous version of sphinx and not the 5prealpha. I should be able to recognize only a small group of words, at most 10, so would be enough for me to adjust some words from Spanish to reach my goal, but if I know from the beginning that I can not recognize the words outside the vocabulary as unknown, I can't use Dialog demo. So do you think it is possible to activate OutOfGrammar probability in dialog demo?

Thanks for your answers, Nikolay. Now, I'm Trying to use the latest sphinx4 release, that you linked in a message ago. I've builded it and the AlignerDemo normally starts. When I start the Dialog Demo I've this error:

Exception in thread "main" java.lang.IllegalStateException: javax.sound.sampled.LineUnavailableException: line with format PCM_SIGNED 16000.0 Hz, 16 bit, mono, 2 bytes/frame, little-endian not supported.
    at edu.cmu.sphinx.api.Microphone.<init>(Microphone.java:38)

I'm using Windows 7, with the latest Java Version. How can I solve this?

I've found that it is a common problem for some users, but I don't understand how to solve it, for run the Dialog Demo.

Thank you very much for your answers, which are ever very useful.