'pocketsphinx-0.7-win32' & 'Germa...

[Bothkill]

I did not succeed in testing the
- 'pocketsphinx-0.7-win32' with
- 'voxforge-de-0.1' (German Acoustic and Language Model).

I've downloaded both packages from this SourceForge project.

I wanted to test the 'test.wav' file provided in the 'voxforge-de-0.1'
package.

I couldn't find info on the parameters which are sent to the
'pocketsphinx_continuous.exe'.

I've looked into the 'voxforge-de-0.1' package and tried to use the command
provided in the 'sphinx3-test' file.
After getting some errors saying that certain files cannot be accessed, I
modified the parameters and almost got it working.

This is the command I used and the error I got in the context of the following
folder structure:

Folder structure:

C:
|--voice
   |--pocketsphinx-0.7-win32
   |--voxforge-de-0.1

Command:

pocketsphinx_continuous.exe ^
 -samprate 8000 ^
 -adcin yes ^
 -cepext ".wav" ^
 -agc none ^
 -adchdr 44 ^
 -ctl "../voxforge-de-0.1/test/test.ctl" ^
 -fsg "../voxforge-de-0.1/test/test.fsg" ^
 -cepdir ./ ^
 -dict "../voxforge-de-0.1/etc/voxforge_de_sphinx.dic" ^
 -fdict "../voxforge-de-0.1/etc/voxforge_de_sphinx.filler" ^
 -bestpath TRUE ^
 -op_mode 2 ^
 -hmm "../voxforge-de-0.1/model_parameters/voxforge_de_sphinx.cd_cont_3000" ^
 -fsgusefiller yes ^
 -infile test.wav

Error (after getting a lot of logging):

ERROR: "fsg_search.c", line 332: The word 'zwei' is missing in the dictionary

But the word 'zwei' is in the .dic file.

If someone has successfully used the German model, please, could you tell me
how to test the 'pocketsphinx-0.7-win32' on the very 'test.wav' file which is
within the 'voxforge-de-0.1' package?
If it's possible, could you give me the exact command to process that file,
assuming I have the above-mentioned folder structure?

Also could someone tell me where to find info on command parameters in order
to use the already built 'pocketsphinx-0.7-win32' package?

And also what is needed to test 'pocketsphinx-0.7-win32' on other German
language audio files (assuming the audio is in the 8000hz 16bit mono format)?
Is the 'voxforge-de-0.1' package enough?

On testing other German language audio files, the program crashes, but this
might be due to me not using the right parameters.

Thank you in advance!

[Bothkill]

One question less to answer, becuase I found some info on the command
parameters at:

http://cmusphinx.sourceforge.net/api/pocketsphinx/

[Nickolay V. Shmyrev]

-samprate 8000

Voxforge German model is 16khz model. You can not use it to recognize 8khz
audio. You need to build 8khz model yourself

If someone has successfully used the German model, please, could you tell me
how to test the 'pocketsphinx-0.7-win32' on the very 'test.wav' file which is
within the 'voxforge-de-0.1' package?

The right command is

pocketsphinx_continuous -hmm ../model_parameters/voxforge_de_sphinx.cd_cont_3000 -dict ../etc/voxforge_de_sphinx.dic -jsgf test.gram  -infile test.wav

But the word 'zwei' is in the .dic file.

Words are case sensitive. zwei in lowercase is not in the dictionary. Grammar
should contain uppercase words.

And also what is needed to test 'pocketsphinx-0.7-win32' on other German
language audio files (assuming the audio is in the 8000hz 16bit mono format)?
Is the 'voxforge-de-0.1' package enough?

No, you need to retrain your own model

[Bothkill]

Thank you vrey, very much nshmyrev!

Thanks to your guidance I succeeded in testing the German language model with
the 'test.wav' file.

Using the '-jsgf test.gram' argument had some results, but not quite accurate.
I used instead '-lm voxforge_de_sphinx.lm' and the accuracy was nearly 100%.

Then I tested a totally different German audio file (16hz mono) and the
program returned three words that had no connection with what was really said
in that audio.
Is it possible to make the program return the 'phones' it detects, without
trying to match any words?

Anyway, I would still need the info on ALL the arguments to pass to
'pocketsphinx-0.7-win32'.
So, running 'pocketsphinx-0.7-win32' with no arguments, gives you all the
arguments (and their current status) the program can receive. But where to
read more about them?
The link I found (my second post) is for the version 0.6 and, for example, it
doesn't say anything on the '-samprate' argument.
I mean, all you guys who now know so much about these things, where did you
get started?

Thank you :)

[Nickolay V. Shmyrev]

Is it possible to make the program return the 'phones' it detects, without
trying to match any words?

Algorithm doesn't detect phones, it looks only for words.

So, running 'pocketsphinx-0.7-win32' with no arguments, gives you all the
arguments (and their current status) the program can receive. But where to
read more about them?

In the speech recognition textbook like

Spoken Language Processing: A Guide to Theory, Algorithm and System
Development
http://www.amazon.com/Spoken-Language-Processing-Algorithm-
Development/dp/0130226165