I've built a Java JNI to sphinx-ii and for the most part it loads up pretty well and all the fbs and utt functions seem to return normally. Can you help me with this problem? I get it to run and I feed it samples from JavaSound (16bit mono pcm 16000hz signed littleendian) for five seconds of sound where I say anything from "sentences" to "corpus" and all I got in the beginning was a recognition of the word "to". Now I get a little more, "to the" or "to the the". Any ideas as to what I'm doing wrong?
For models, I'm using all the files that were distributed with the package from sourceforge. The following is the console output from my program (which I can explain if needed). Let me know also if you want the source code or compiled binary asr.dll:
kb() attempting to read phone file C:\Projects\sphinx2-0.3\model\hmm\6k\phone
phone_read() attempting to read phone file C:\Projects\sphinx2-0.3\model\hmm\6k\
phone
phone_read() return value is 0
init called with result 0
Result of start request: true
beginUtt called (ignoring name passed)
beginUtt result was 0 before java conversion
Result of begin utterance request: true
addData called with 80640 samples
the result of uttproc_rawdata was 0
addData returned: true
Java is about to call the get result function
getResult called
End utterance processing returned 0
Attempting to get result....
@Frm 26:
@Frm 58: TO
@Frm 90: TO THE
@Frm 122: TO THE
@Frm 154: TO THE
@Frm 186: TO THE
@Frm 218: TO THE
@Frm 250: TO THE
@Frm 282: TO THE
@Frm 314: TO THE
@Frm 346: TO THE THE
@Frm 378: TO THE THE
@Frm 410: TO THE THE
@Frm 442: TO THE THE
@Frm 474: TO THE THE
FINAL RESULT @frm -4: TO
Process returned 0
Process result returned with -4 frames
Process result returned recstring: TO
Java got a result of: true
Stop called
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2001-07-29
I'm interested in your Java code. If you make it available, I'll work to help you solve your problems.
Thanks, zach
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2001-08-09
It sounds like you might be using the language model and pronunciation dictionary from the turtle demo (demo.lm and demo.dic). Check the arguments you are using with the -lmfn and -dictfn flags.
Sphinx will only recognize words that appear both in the pronunciation dictionary and language model. If you are using the files from the turtle demo, neither "corpus" nor "sentences" can be recognized because those words do not appear in the turtle demo files. For a quick test, you could test the system's recognition only on words that appear in the demo.corpus file, which was used to create the demo.lm and demo.dic files. If you want to use Sphinx with a different vocabulary, you will need to create a new language model and pronunciation dictionary.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I've built a Java JNI to sphinx-ii and for the most part it loads up pretty well and all the fbs and utt functions seem to return normally. Can you help me with this problem? I get it to run and I feed it samples from JavaSound (16bit mono pcm 16000hz signed littleendian) for five seconds of sound where I say anything from "sentences" to "corpus" and all I got in the beginning was a recognition of the word "to". Now I get a little more, "to the" or "to the the". Any ideas as to what I'm doing wrong?
For models, I'm using all the files that were distributed with the package from sourceforge. The following is the console output from my program (which I can explain if needed). Let me know also if you want the source code or compiled binary asr.dll:
kb() attempting to read phone file C:\Projects\sphinx2-0.3\model\hmm\6k\phone
phone_read() attempting to read phone file C:\Projects\sphinx2-0.3\model\hmm\6k\ phone
phone_read() return value is 0
init called with result 0
Result of start request: true
beginUtt called (ignoring name passed)
beginUtt result was 0 before java conversion
Result of begin utterance request: true
addData called with 80640 samples
the result of uttproc_rawdata was 0
addData returned: true
Java is about to call the get result function
getResult called
End utterance processing returned 0
Attempting to get result....
@Frm 26:
@Frm 58: TO
@Frm 90: TO THE
@Frm 122: TO THE
@Frm 154: TO THE
@Frm 186: TO THE
@Frm 218: TO THE
@Frm 250: TO THE
@Frm 282: TO THE
@Frm 314: TO THE
@Frm 346: TO THE THE
@Frm 378: TO THE THE
@Frm 410: TO THE THE
@Frm 442: TO THE THE
@Frm 474: TO THE THE
FINAL RESULT @frm -4: TO
Process returned 0
Process result returned with -4 frames
Process result returned recstring: TO
Java got a result of: true
Stop called
I'm interested in your Java code. If you make it available, I'll work to help you solve your problems.
Thanks, zach
It sounds like you might be using the language model and pronunciation dictionary from the turtle demo (demo.lm and demo.dic). Check the arguments you are using with the -lmfn and -dictfn flags.
Sphinx will only recognize words that appear both in the pronunciation dictionary and language model. If you are using the files from the turtle demo, neither "corpus" nor "sentences" can be recognized because those words do not appear in the turtle demo files. For a quick test, you could test the system's recognition only on words that appear in the demo.corpus file, which was used to create the demo.lm and demo.dic files. If you want to use Sphinx with a different vocabulary, you will need to create a new language model and pronunciation dictionary.
Can you contact me? I'd like to take a look, and make sure this gets working.
lenzo@cs.cmu.edu
kevin