Acoustic Model Discussions

Flat
Best way for my chatterbot application
User: mendroid
Date: 12/31/2009 8:32 am
Views: 5753
Rating: 6

Hi again,

finally, i finished an applet to communicate my chatterbot with Julius. With the examples from Julius Install: "Call Ken" "Phone Ken" ....., etc my chatterbot is running perfectly. 

But, now i want to make my own sentences like:

Hello machine

What time is it?

What is your name?

.....

Which is the best way and easy to training Julius for "short sentences". Note that is a chatterbot and i dont need "continuos speeching". Im very confuse with the grammar & voca files. I need these files for short sentences?

Thank you in advance for your help.

Mendru

--- (Edited on 12/31/2009 8:32 am [GMT-0600] by mendroid) ---

Re: Best way for my chatterbot application
User: kmaclean
Date: 1/5/2010 1:09 pm
Views: 148
Rating: 5

>Which is the best way and easy to training Julius for "short sentences".

Step 1 one of the VoxForge tutorial provides an explanation of how the Julius/Julian grammar works.

--- (Edited on 1/5/2010 2:09 pm [GMT-0500] by kmaclean) ---

Re: Best way for my chatterbot application
User: mendroid
Date: 1/8/2010 3:36 am
Views: 87
Rating: 8

Thank you so much Kmaclean for your answer.

Now, im testing your last Acoustic Model files with my "basic" .grammar & .voca files & my chatterbot. With first tests the accuracy is near 95% with my voice WOW !!.

What do you think about Omnidirectional MICs and Unidirectional MICs? ( button mics ). Which is the best for use with Julius?

 

Thank you so much!!!

Mendru

--- (Edited on 1/8/2010 3:36 am [GMT-0600] by mendroid) ---

Re: Best way for my chatterbot application
User: kmaclean
Date: 1/8/2010 8:24 am
Views: 150
Rating: 5

>What do you think about Omnidirectional MICs and Unidirectional

>MICs? ( button mics ). Which is the best for use with Julius?

Don't know...

If you can try both out and let us know of your experiences, it would be greatly appreciated.

Ken

--- (Edited on 1/8/2010 9:24 am [GMT-0500] by kmaclean) ---

Re: Best way for my chatterbot application
User: mendroid
Date: 1/12/2010 3:19 am
Views: 179
Rating: 6

Hi Ken, me again...

i finished my test with button mics. I dont know why the best accuracy is with Ominidirectional Mic and not with Unidirectional Mic ( noise reduction ) but the difference is Omni: Accuracy ( 90-95% ) and Uni with Noise reduction: Accuracy ( 80-85% ). Very strange results!!!

Ken, how can i test your last Acoustic Model without .grammar & .voca files. Only the acoustic model for continuos dictation.

Another question is ( sorry for my english ). When a phrase is not 100% recognized by Julius and change for example "now are you?" instead "how are you?", my chatterbot fail with the answer. Is it possible to put a recognition level accuracy in Julius?. For example, If the phrase recognized by Julius is under 70%, dont shot it and listen again. ( Sorry again for my english ). I hope that you understand my question.

Thanks in advance Ken,

Mendru

--- (Edited on 1/12/2010 3:19 am [GMT-0600] by mendroid) ---

Re: Best way for my chatterbot application
User: kmaclean
Date: 1/19/2010 1:13 pm
Views: 102
Rating: 8

>Is it possible to put a recognition level accuracy in Julius?. For example,

>If the phrase recognized by Julius is under 70%, dont shot it and listen

>again.

Not unless you want to change Julius... regardless you can do this at your application level by looking at the confidence scores (my post at the end).

Ken

--- (Edited on 1/19/2010 2:13 pm [GMT-0500] by kmaclean) ---

Re: Best way for my chatterbot application
User: kmaclean
Date: 1/19/2010 1:15 pm
Views: 2266
Rating: 6

>Ken, how can i test your last Acoustic Model without .grammar & .voca

>files. Only the acoustic model for continuos dictation.

See this post: Continuous speech dictation with lm_giga_64k_nvp_3gram

Ken

--- (Edited on 1/19/2010 2:15 pm [GMT-0500] by kmaclean) ---

PreviousNext