I have downloaded Sphinx2 few days ago, compiled it. All worked fine.
But when I launch the sphinx2-demo binary, I can't get any recognition. Most of the time, I just get [silence] [audio]
and nothing more.
If I shout, it seems to react but the answer it gives me is completly wrong.
What's the matter with it???
Is there some config file that I missed ???
I'm using the alsa-driver with OSS emulation. Maybe it is the reason of my problems ?
Any help greatly appreciated.
Cheers,
Tony
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Note that alsa 1.5.0 and later won't work right with sphinx. Are you using the demo language models or did you create your own? If you are using the demo model's the vocab is very strange. You might consider creating your own. I am also new to this, but I think I'm correct on this one, hope rather... Good luck...
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2000-03-26
What do you mean by won't work right ??? I am using the alsa driver version 0.5.5 and it does seem to work expect that it can't recognise any word I say. I also try with the free OSS driver given with the Slakware 7.0 and the result is the same.
What will change if I use my own language ???
Regards.
Tony.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm not sure about the alsa issue. I read somewhere on the site that some alsa does not work correctly with sphinx due to a change in the api. That's all I know about that...
As for the language model...
If you create your own language model you can program certain words in that you want it to recognize. The demo model doesn't contain many regularly used words. That might be part of your problem. Does it even recognize "HELLO?"
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2000-03-26
I think (but I'm not sure) the recogniser initialises the soundcard mixer with its own values which are too low for my configuration. That result is that I am oblige to shout to be "heard" and this modifies the way I speak so that the recogniser can't understand what I'm saying.
Can anyone confirm that ???
I really need to make his software working.
Cheers,
Tony.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The ALSA drivers set the volumes to ZERO when they are loaded by default, and then again later when they are autocleaned from the kernel (and possibly when reloaded).
Check with xmixer or gamix that your master volume, mic volume, and input gain (igain) are at about 80%. Try recording some audio (using arecord) and playing it back to make sure it doesn't sound too distorted. If you can set your mixer so that you can hear the microphone input through the headphones, you can get a good idea if it's working, but still make sure that the igain is up.
Also, try out sphinx2-test and make sure it decodes the example utterance 'goforward' as "GO FORWARD TEN METERS".
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi everybody...
I have downloaded Sphinx2 few days ago, compiled it. All worked fine.
But when I launch the sphinx2-demo binary, I can't get any recognition. Most of the time, I just get [silence] [audio]
and nothing more.
If I shout, it seems to react but the answer it gives me is completly wrong.
What's the matter with it???
Is there some config file that I missed ???
I'm using the alsa-driver with OSS emulation. Maybe it is the reason of my problems ?
Any help greatly appreciated.
Cheers,
Tony
Note that alsa 1.5.0 and later won't work right with sphinx. Are you using the demo language models or did you create your own? If you are using the demo model's the vocab is very strange. You might consider creating your own. I am also new to this, but I think I'm correct on this one, hope rather... Good luck...
What do you mean by won't work right ??? I am using the alsa driver version 0.5.5 and it does seem to work expect that it can't recognise any word I say. I also try with the free OSS driver given with the Slakware 7.0 and the result is the same.
What will change if I use my own language ???
Regards.
Tony.
I'm not sure about the alsa issue. I read somewhere on the site that some alsa does not work correctly with sphinx due to a change in the api. That's all I know about that...
As for the language model...
If you create your own language model you can program certain words in that you want it to recognize. The demo model doesn't contain many regularly used words. That might be part of your problem. Does it even recognize "HELLO?"
The tool to create a language model is at:
http://alf14.speech.cs.cmu.edu:8044/lmtool.html
c0re
I think (but I'm not sure) the recogniser initialises the soundcard mixer with its own values which are too low for my configuration. That result is that I am oblige to shout to be "heard" and this modifies the way I speak so that the recogniser can't understand what I'm saying.
Can anyone confirm that ???
I really need to make his software working.
Cheers,
Tony.
The ALSA drivers set the volumes to ZERO when they are loaded by default, and then again later when they are autocleaned from the kernel (and possibly when reloaded).
Check with xmixer or gamix that your master volume, mic volume, and input gain (igain) are at about 80%. Try recording some audio (using arecord) and playing it back to make sure it doesn't sound too distorted. If you can set your mixer so that you can hear the microphone input through the headphones, you can get a good idea if it's working, but still make sure that the igain is up.
Also, try out sphinx2-test and make sure it decodes the example utterance 'goforward' as "GO FORWARD TEN METERS".