speech-tools/grammar/wfst/wfst_transduce.cc

324 lines
9.6 KiB
C++
Raw Permalink Normal View History

2015-09-19 10:52:26 +02:00
/*************************************************************************/
/* */
/* Centre for Speech Technology Research */
/* University of Edinburgh, UK */
/* Copyright (c) 1997 */
/* All Rights Reserved. */
/* */
/* Permission is hereby granted, free of charge, to use and distribute */
/* this software and its documentation without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of this work, and to */
/* permit persons to whom this work is furnished to do so, subject to */
/* the following conditions: */
/* 1. The code must retain the above copyright notice, this list of */
/* conditions and the following disclaimer. */
/* 2. Any modifications must be clearly marked as such. */
/* 3. Original authors' names are not deleted. */
/* 4. The authors' names are not used to endorse or promote products */
/* derived from this software without specific prior written */
/* permission. */
/* */
/* THE UNIVERSITY OF EDINBURGH AND THE CONTRIBUTORS TO THIS WORK */
/* DISCLAIM ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING */
/* ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT */
/* SHALL THE UNIVERSITY OF EDINBURGH NOR THE CONTRIBUTORS BE LIABLE */
/* FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES */
/* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN */
/* AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, */
/* ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF */
/* THIS SOFTWARE. */
/* */
/*************************************************************************/
/* Author : Alan W Black */
/* Date : December 1997 */
/*-----------------------------------------------------------------------*/
/* */
/* Transduction using a WFST */
/* */
/*=======================================================================*/
#include <iostream>
#include "EST_WFST.h"
/** An internal class in transduction of WFSTs holding intermediate
state information.
*/
class wfst_tstate {
public:
int state;
EST_IList outs;
float score;
};
ostream &operator << (ostream &s, const wfst_tstate &state)
{
(void)state;
return s << "<<wfst_tstate>>";
}
Declare_TList(wfst_tstate)
#if defined(INSTANTIATE_TEMPLATES)
#include "../base_class/EST_TList.cc"
Instantiate_TList(wfst_tstate)
#endif
typedef EST_TList<wfst_tstate> wfst_tstate_list;
static void add_transduce_mstate(const EST_WFST &wfst,
const wfst_tstate &cs,
wfst_translist &tranlist,
wfst_tstate_list &ns);
int transduce(const EST_WFST &wfst,const EST_StrList &in,EST_StrList &out)
{
// Map names to internal ints before transduction
EST_Litem *p;
EST_IList in_i,out_i;
int r;
for (p=in.head(); p != 0; p=p->next())
in_i.append(wfst.in_symbol(in(p)));
r = transduce(wfst,in_i,out_i);
for (p=out_i.head(); p != 0; p=p->next())
out.append(wfst.out_symbol(out_i(p)));
return r;
}
int transduce(const EST_WFST &wfst,const EST_IList &in,EST_IList &out)
{
// Transduce input stream to an output stream
EST_Litem *i,*cs;
int r=FALSE;
wfst_tstate_list *current_ms = new wfst_tstate_list;
wfst_tstate start_state;
wfst_translist ss_eps_trans;
start_state.state = wfst.start_state();
start_state.score = 0;
current_ms->append(start_state);
// Add any epsilon accessible states
wfst.transduce(wfst.start_state(),wfst.in_epsilon(),ss_eps_trans);
add_transduce_mstate(wfst,start_state,ss_eps_trans,*current_ms);
for (i=in.head(); i != 0; i=i->next())
{
wfst_tstate_list *ns = new wfst_tstate_list;
for (cs=current_ms->head(); cs != 0; cs=cs->next())
{ // For each state in current update list of new states
wfst_translist translist;
wfst.transduce((*current_ms)(cs).state,in(i),translist);
add_transduce_mstate(wfst,(*current_ms)(cs),translist,*ns);
}
// Using pointers to avoid having to copy the state list
delete current_ms;
current_ms = ns;
if (current_ms->length() == 0)
break; // give up, no transition possible
}
// current_ms will contain the list of possible transitions
if (current_ms->length() > 1)
cerr << "WFST: found " << current_ms->length() << " transductions" <<
endl;
// should find "best" but we'll find longest at present
// Choose the longest (should be based on score)
for (cs = current_ms->head(); cs != 0; cs=cs->next())
{
if ((wfst.final((*current_ms)(cs).state)) &&
((*current_ms)(cs).outs.length() > out.length()))
{
r = TRUE;
out = (*current_ms)(cs).outs;
}
}
delete current_ms;
return r;
}
static void add_transduce_mstate(const EST_WFST &wfst,
const wfst_tstate &cs,
wfst_translist &translist,
wfst_tstate_list &ns)
{
// For each possible transduction of in from cs.state in WFST
// add it to ns.
// This should really only add states if they are not already there
// but you really need stae plus recognized path to get all
// transductions so a tree structure would be better.
EST_Litem *t;
// Add new states to ns if not already there
for (t=translist.head(); t != 0; t=t->next())
{
// Declare a new one and put it on the end of the list
// before we fill its values, this saves a copy
wfst_tstate tts;
ns.append(tts);
wfst_tstate &ts = ns.last();
ts.state = translist(t)->state();
// the combination is probably WFST dependant
ts.score = translist(t)->weight()+cs.score;
// copying the outs up to now (pity)
ts.outs = cs.outs;
ts.outs.append(translist(t)->out_symbol());
// Also any potential epsilon transitions for this new state
wfst_translist etranslist;
wfst.transduce(ts.state,wfst.in_epsilon(),etranslist);
add_transduce_mstate(wfst,ts,etranslist,ns);
}
}
int recognize(const EST_WFST &wfst,const EST_StrList &in,int quiet)
{
// Map names to internal ints before recognition
EST_Litem *p;
EST_IList in_i,out_i;
int i,o;
int r;
for (p=in.head(); p != 0; p=p->next())
{
if (in(p).contains("/"))
{
i = wfst.in_symbol(in(p).before("/"));
o = wfst.out_symbol(in(p).after("/"));
}
else
{
i = wfst.in_symbol(in(p));
o = wfst.out_symbol(in(p));
}
in_i.append(i);
out_i.append(o);
}
r = recognize(wfst,in_i,out_i,quiet);
return r;
}
int recognize(const EST_WFST &wfst,const EST_IList &in,
const EST_IList &out, int quiet)
{
int state = wfst.start_state();
EST_Litem *p,*q;
int nstate;
for (p=in.head(),q=out.head();
((p != 0) && (q != 0));
p=p->next(),q=q->next())
{
nstate = wfst.transition(state,in(p),out(q));
if (!quiet)
printf("state %d %s/%s -> %d\n",state,
(const char *)wfst.in_symbol(in(p)),
(const char *)wfst.out_symbol(out(q)),
nstate);
state = nstate;
if (state == WFST_ERROR_STATE)
return FALSE;
}
if (p != q)
{
cerr << "wfst recognize: in/out tapes of different lengths"
<< endl;
return FALSE;
}
if (wfst.final(state))
return TRUE;
else
return FALSE;
}
int recognize_for_perplexity(const EST_WFST &wfst,
const EST_StrList &in,
int quiet,
float &count,
float &sumlogp)
{
// Map names to internal ints before recognition
EST_Litem *p;
EST_IList in_i,out_i;
int i,o;
int r;
for (p=in.head(); p != 0; p=p->next())
{
if (in(p).contains("/"))
{
i = wfst.in_symbol(in(p).before("/"));
o = wfst.out_symbol(in(p).after("/"));
}
else
{
i = wfst.in_symbol(in(p));
o = wfst.out_symbol(in(p));
}
in_i.append(i);
out_i.append(o);
}
r = recognize_for_perplexity(wfst,in_i,out_i,quiet,count,sumlogp);
return r;
}
int recognize_for_perplexity(const EST_WFST &wfst,
const EST_IList &in,
const EST_IList &out,
int quiet,
float &count,
float &sumlogp)
{
int state = wfst.start_state();
EST_Litem *p,*q;
int nstate;
float prob;
count = 0;
sumlogp = 0;
for (p=in.head(),q=out.head();
((p != 0) && (q != 0));
p=p->next(),q=q->next())
{
nstate = wfst.transition(state,in(p),out(q),prob);
count++;
if (prob > 0)
sumlogp += log(prob);
else
sumlogp += -100; // bad hack
if (!quiet)
printf("state %d %s/%s -> %d\n",state,
(const char *)wfst.in_symbol(in(p)),
(const char *)wfst.out_symbol(out(q)),
nstate);
state = nstate;
if (state == WFST_ERROR_STATE)
return FALSE;
}
if (p != q)
{
cerr << "wfst recognize: in/out tapes of different lengths"
<< endl;
return FALSE;
}
if (wfst.final(state))
return TRUE;
else
return FALSE;
}