Hi, I've modified livedecode on windows xp to loop after each 5 second burst of speaking. It still waits for input and quits perfectly fine if you enter q. The issue I'm having is the first time through, it correctly recognizes the silence (no microphone plugged in), but the second time through I get:
"Re-normalizing the previous score"
several times and what should be silence interpreted as "++COUGH++". The program then exits with an error (windows caught) which suggests either an invalid memory allocation, or access to memory that is no longer being used.
I've played around with it quite a bit, tried freeing memory and re-allocating it, thinking it was trying to read something that was now junk, but I came up with nice errors (xp based errors, my guess is invalid memory access) or no change in behavior.
I'd be happy if someone could even point me to what I need to re-initialize for each loop.
Thanks,
Matt
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Two issues:
-Sphinx is mainly a speech recognizer, it is only a reasonable end-pointer. You can even say it is a poor noise detector. At the same time, you got to realize everytime you say something, even you say the same words or even you thought you say the same thing, the waveforms are totally different.
-Sphinx 3 livedecode is a demo. It will only record 5 second of speech and try to do recognition according to the grammar of the demo. The goal of Sphinx3 is mainly giving a set of APIs for developers.
The core dump you reported though, was known to me for a while. I will try to fix it in next release.
If you need to see another implementation using the live-mode APIs, please take a look of main_livepretend.c
Arthur
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi, I've modified livedecode on windows xp to loop after each 5 second burst of speaking. It still waits for input and quits perfectly fine if you enter q. The issue I'm having is the first time through, it correctly recognizes the silence (no microphone plugged in), but the second time through I get:
"Re-normalizing the previous score"
several times and what should be silence interpreted as "++COUGH++". The program then exits with an error (windows caught) which suggests either an invalid memory allocation, or access to memory that is no longer being used.
I've played around with it quite a bit, tried freeing memory and re-allocating it, thinking it was trying to read something that was now junk, but I came up with nice errors (xp based errors, my guess is invalid memory access) or no change in behavior.
I'd be happy if someone could even point me to what I need to re-initialize for each loop.
Thanks,
Matt
Two issues:
-Sphinx is mainly a speech recognizer, it is only a reasonable end-pointer. You can even say it is a poor noise detector. At the same time, you got to realize everytime you say something, even you say the same words or even you thought you say the same thing, the waveforms are totally different.
-Sphinx 3 livedecode is a demo. It will only record 5 second of speech and try to do recognition according to the grammar of the demo. The goal of Sphinx3 is mainly giving a set of APIs for developers.
The core dump you reported though, was known to me for a while. I will try to fix it in next release.
If you need to see another implementation using the live-mode APIs, please take a look of main_livepretend.c
Arthur