Thank you nsh,

look, I am recording my own voice into a sph file with this command:

arecord cen1-mjasp-a.sph -f S16_LE -c1 -r16500 -d5 -traw

Then I put the files inside wav/an4test_clstk/mjap
plus this:
echo "an4test_clstk/mjap/cen1-mjap-a" >> etc/an4_test.fileid

Then, when I do
perl scripts_pl/make_feats.pl  -ctl etc/an4_test.fileids

It is running for more than 45 minutes in a dual core machine.

I cancelled

Am I doing something wrong?

Is there a way to setup it more close to real time recognition?

--- (Edited on 10/13/2009 1:30 pm [GMT-0500] by Visitor) ---

> Am I doing something wrong?

Yes, you are trying to put raw audio into sph file. That's definitely something 
system doesn't expect.

--- (Edited on 10/14/2009 01:10 [GMT+0400] by nsh) ---

Hum, right, make sense.

I did like this because analyzing the sph file seamed to be raw.

I don't know exactlly what is a sph file, do you have a hint about how could I record it? what program could I use to produce a sph file?

--- (Edited on 10/14/2009 3:25 am [GMT-0500] by Visitor) ---

There is no need to record sph files. Database can contain wav files. You need to change the file type in configuration in etc/sphinx_train.cfg:

# Audio waveform and feature file information
$CFG_WAVFILES_DIR = "$CFG_BASE_DIR/wav";
$CFG_WAVFILE_EXTENSION = 'wav';
$CFG_WAVFILE_TYPE = 'mswav'; # one of nist, mswav, raw
$CFG_FEATFILES_DIR = "$CFG_BASE_DIR/feat";
$CFG_FEATFILE_EXTENSION = 'mfc';

--- (Edited on 10/14/2009 12:33 [GMT+0400] by nsh) ---

Thanks a lot nsh,

It works smothlly, It was able to create the features & decode my sentence in a few seconds.

Of course I got some missdecoding spetially because I didn't do the learning by myself yet, I am using an4 learning dataset.

Next step now I will create my own dataset for learning.

--- (Edited on 10/14/2009 4:49 am [GMT-0500] by Visitor) ---

When I perform a decode, I can check the words in log file.

Is there a way to know which phones are being generating?

Since there are more than one combination of phones per word in dictionary I don't know which one is being ellected.

--- (Edited on 10/14/2009 10:50 am [GMT-0500] by Visitor) ---

-backtrace yes does this for example

--- (Edited on 10/15/2009 02:04 [GMT+0400] by nsh) ---

Phase 4: CTL - Checking number of lines in the transcript should match lines in control file

FAILED

what is the control file?

--- (Edited on 10/15/2009 11:15 am [GMT-0500] by Visitor) ---

why I cannot succeed with training new words in the dictionary? only the words existant in the previous dictionary are working, the new ones are never catch up

--- (Edited on 10/15/2009 4:29 pm [GMT-0500] by Visitor) ---

Hello Nsh,

I looked in the "../logdir/decode/an4-1-1.log" but all i see is "Tue Mar 23 01:45:42 2010". There is nothing in there that tells me what error to look for. I am using a 32 bit vista machine and I would appreciate some help. Thank you for your time.

--- (Edited on 3/23/2010 2:07 am [GMT-0500] by Visitor) ---