Recognition accuracy is lower than accuracy of acoustic model

[wunna]

Dear All members,
I used the following code to recognitze the recording file using StreamSpeakRecognizer.

Configuration configuration = new Configuration();

        configuration
                .setAcousticModelPath("file:///D:/speaker_dependent/10th/wunna10/model_parameters/wunna10.cd_cont_200");
        configuration
                .setDictionaryPath("file:///D:/speaker_dependent/10th/wunna10/etc/wunna10.dic");
        configuration
                .setLanguageModelPath("file:///D:/speaker_dependent/10th/wunna10/etc/wunna10.lm.dmp");

        StreamSpeechRecognizer recognizer = new StreamSpeechRecognizer(configuration);
        InputStream stream = new FileInputStream(new File("D:/WordAm/fifthtest.wav"));

        recognizer.startRecognition(stream);
        SpeechResult result;
        while ((result = recognizer.getResult()) != null) {
            System.out.format("Hypothesis: %s\n", result.getHypothesis());
        }
        recognizer.stopRecognition();

    }

The accuracy of this recognition system is lower than the accuracy of acoustci model. Is it the normal? The accuracy of my acoustic model gets 83%, but the accuracy of that recognition program is lower. The result text of this recognition program is always wrong in the beginning of the recording files. My acoustic model can be checked in the following link.https://www.dropbox.com/s/ha5gchhbyy0puwx/wunna10.zip?dl=0

[wunna]

Hi
I am no experience in speech recognition. I have trained my acoustic model for Myanmar. WER of this AM is 16 and its accuracy is over 83. I used that AM in speech recognition code with Sphinx4.When I run that recognition program, some recording files can be recognized and some files cause java heap space error. I solved that problem with -Xms and -Xmx parameter values. But this error still be. my acousit model can be downloaded in the above post and my code is there. Please don't skip my post. And please guide me how to solve that problem. I used 64 bit window 7, core i5 and 4G memory laptop for recognition.