You need to write your own postprocessing code that will do that. For example
it might be fifty seven or seventy eight. You need to analyze the text and
convert it to 57 or 78.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
this is insane, there must be something to translate the text number in
number. Something like that, if you want the number, the grammar (fsg) looks
like that:
I haven't used pocketsphinx but in sphinx you can write dictionary words which
are actually numbers, as long as they precede with an underscore. e.g. _5 ,
_58
You can strip off _ from decoder output to get what you want.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi my question is the next.
when i speak and pocketsphinx return the hyp, it is like that:
ok, i need that:
it is posible convert
five in 5??
i am ussing fsg like this:
Hello
You need to write your own postprocessing code that will do that. For example
it might be fifty seven or seventy eight. You need to analyze the text and
convert it to 57 or 78.
But..... I have proces 100 numbers?? like that?
I suppose that cadena have the text number
really?
with all numbers?
this is insane, there must be something to translate the text number in
number. Something like that, if you want the number, the grammar (fsg) looks
like that:
This isn't work, i need some help please, any idea?
I haven't used pocketsphinx but in sphinx you can write dictionary words which
are actually numbers, as long as they precede with an underscore. e.g. _5 ,
_58
You can strip off _ from decoder output to get what you want.
So thanks!! I don't know that.
I try and coment soon :D
It depends how you will implement this. Reasonable implementation shouldn't
take more than 100 lines of code
Yes, but nobody contributed this part yet
This will not work great with things like "fifty eigth" which will be "_50 _8"
and you need a postprocessing anyway.
For example how is it done in real dialog systems you can check Olympus
http://wiki.speech.cs.cmu.edu/olympus/index.php/DateTime
Which has all functionality implemented. The only issue is that it's a huge
software package doing many things beside the specified task.
hi! I try to write
<digit> = one_1
| two_2
| three_3
...
;
But it doesn't work :( </digit>
ERROR: "fsg_search.c", line 332: The word '_1' is missing in the dictionary
and, I am working on one robot calls HOAP-3 and I havn't more free space to
install Olympus,
what must I do?
Any ideas?? Please help me!!! :(
Hi,
Words in FSG need to exactly match with words in dictionary, otherwise such
errors will come. Which words are there in your dictionary?
well, in the dic I have this:
turtle.dic
match with the default dic:
do you recomend that I must change??
I only want to reconoce this sentences:
only this and always in this order, Sometimes auto-recognition has wrong hyp
like:
hello robot five bye step
or something like that and I can't fix.
What must I do to improve the accuracy?
Now, I program the translation of the numbers like that:
if someone like it, take it, It works fine.
ups, I have this in my program, It is weird to run?
config = cmd_ln_init(NULL, ps_args(), TRUE,
"-hmm", MODELDIR "/hmm/en_US/hub4wsj_sc_8k",
"-lm", MODELDIR "/lm/en/turtle.DMP",
"-jsgf","/usr/share/pocketsphinx/grammar/HoapGrammar.jsgf",NULL);
I don't have .dic indeed.
I must load?? or with my jsgf run?
You need to specify either language model or jsgf grammar. There is no sense
to specify both. Remove
part
You need to specify a dictionary
To improve accuracy you need to measure it first, you can find some
information on that in the tutorial
http://cmusphinx.sourceforge.net/wiki/tutorialam
This is a nice piece of code, it would be cool to make it a part of
pocketsphinx. Can you make a patch for pocketsphinx?