yunyin / transducersaurus Goto Github PK

Hi,

on my system (Ubuntu 9.04, g++ 4.3.3) g++ complains about not being able to 
find the 'dlopen' and 'dlerror'. The following trivial change fixes it:

diff -r 7dc2847835c3 cpp/compile.sh
--- a/cpp/compile.sh    Sun May 22 12:31:20 2011 +0900
+++ b/cpp/compile.sh    Sun May 22 12:27:48 2011 +0300
@@ -1,3 +1,3 @@
 #!/bin/bash
 #Compile the normalizer
-g++ -O2 -o normalizeG normalizeG.cpp -lfst
\ No newline at end of file
+g++ -O2 -o normalizeG normalizeG.cpp -lfst -ldl
\ No newline at end of file

The full cascade optimization operations should work, but the resulting cascade 
will not have the correct input symbols and will require manual removal of the 
input aux symbols and anything else like failure transitions, etc.  The HTK AM 
variant already has a 'mapper' that will perform this job as a final step; need 
to write one for the sphinx format cascades...

It would be nice to have a proper python module that we can install, rather 
than having to create a sandbox.

The regex compiler is still incomplete/incorrect.
Best not to use it yet.

Would be nice to add this. No dependency on sphinx scripts, just a pure text 
analyzer of the mdef file.

It would be nice to have a debian package of the python module (assuming the 
python module gets built!)

This grammar doesn't work properly:

CITY1 := BOSTON
CITY10 := CHICAGO
result := ( $CITY1 | CITY10 )

The rule CITY10 got replaced with CITY1 + 0

The fix is attached

i broke the am mapping for htk and sphinx models. i done goofed. 

If phoneset is large it can become enourmously big. Here is the patch to reduce 
some unneeded transitions

The current components generation scripts have hard-wired command line args and 
no help message.  Should build proper help and arg parsing routines for the 
__main__ functions.

All log messages are currently generated with 'print' statements.  This is not 
so cool and we should really be using the 'logging' module to do this properly. 
 Anyone want to fix it up?!?

@@ -227,7 +252,7 @@
             self._make_arc( self.start, self.eps, lp )
             self._make_aux( self.eps, lp )
             #Monophone arcs
-            self._make_arc( self.sil, lp, self.eps )
+            self._make_arc( self.eps, lp, self.eps )
             self._make_final( lp, self.eps )
             for mp in self.phons:
                 #Initial to Internal arcs

It should be eps instead of sil.