It looks like you are using pocketsphinx trunk that contains a bug that broke decoding. Could you please try released version 0.4.1. You'll also need to revert sphinxbase.
Also please share the result file /usr/local/src/tutorial/an4/result/an4-1-1.match to check what's broken there. We'll try to fix the bug.
--- (Edited on 6/26/2008 2:25 am [GMT-0500] by nsh) ---
Hi,
I tried pocketsphinx released version 0.4.1 and reverted to released version of sphinxbase. This time it seems decoding was done, but got 100% error.
The log file is here: http://sip.hq.com.au/an4-1-1.log.new
and the result file is here: http://sip.hq.com.au/an4-1-1.match
Please help.
Kind Regards,
Zia
--- (Edited on 7/1/2008 9:20 pm [GMT-0500] by akmzia) ---
--- (Edited on 7/2/2008 3:10 am [GMT-0500] by nsh) ---
Thanks for the quick reply.
All the files in the an4 folder are here:
If you want to download the the entire folder, u can download the tar from this link:
Thanks again,
Zia
--- (Edited on 7/2/2008 7:10 am [GMT-0500] by akmzia) ---
--- (Edited on 7/2/2008 7:13 am [GMT-0500] by akmzia) ---
Thanks, I looked closer on this. There are multiple issues here:
1. With pocketsphinx-0.4 I can decode just fine with your models:
SENTENCE ERROR: 53.077% (69/130)
But you can't decode as I see. There are multiple issues possible here:
a) pocketsphinx-0.4 has 64-bit issues (most likely)
b) You are using pocketsphinx-0.4 but sphinxbase from trunk somehow (less likely)
2. I also tried trunk and there is the following issue. It decodes just fine, but unforunately it dumps result in a different format:
RUBOUT G N EIGHTY M THREE NINTH (an4test_clstk/fcaw/an406-fcaw-b -78722380)
instead of
RUBOUT T N EIGHTY M THREE NINE (an406-fcaw-b -78761223)
so scripts fail to compare strings. To fix it I suggest the following patch:
Index: slave.pl
===================================================================
--- slave.pl (revision 8523)
+++ slave.pl (working copy)
@@ -111,7 +111,16 @@
m/^(.*)\((\S+)\)\s*$/;
# Make them uppercase
my $text = uc($1);
- my $id = uc($2);
+
+# Select basename from uttid
+ @path = split /[\\\/]/, uc(2);
+ my $id;
+ if ($#path > 0) {
+ $id = $path[$#path - 1];
+ } else {
+ $id = "id";
+ }
+
# Removing leading spaces
$text =~ s/^\s+//;
# Removing trailing spaces
To remove directory path from uttid and revert back to previous format. But I don't think it's correct, so it needs review to be applied to trunk. You can try it and you should get the same result with trunk as with 0.4. If not, please upload again the decoding log file.
--- (Edited on 7/2/2008 4:01 pm [GMT-0500] by nsh) ---