#!/bin/bash
# Copyright 2012-2013  Johns Hopkins University (Author: Daniel Povey);
#                      Arnab Ghoshal
#                2014  Guoguo Chen
#                2015  Hainan Xu
#                2016  FAU Erlangen (Author: Axel Horndasch)
#                2019  Dongji Gao

# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#  http://www.apache.org/licenses/LICENSE-2.0
#
# THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED
# WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,
# MERCHANTABLITY OR NON-INFRINGEMENT.
# See the Apache 2 License for the specific language governing permissions and
# limitations under the License.

# This script prepares a directory (for subword) such as data/lang_subword/, in the standard format,
# given a source directory containing a subword dictionary lexicon.txt in a form like:
# subword phone1 phone2 ... phoneN
# per line (alternate prons would be separate lines), or a dictionary with probabilities
# called lexiconp.txt in a form:
# subword pron-prob phone1 phone2 ... phoneN
# (with 0.0 < pron-prob <= 1.0); note: if lexiconp.txt exists, we use it even if
# lexicon.txt exists.
# and also files silence_phones.txt, nonsilence_phones.txt, optional_silence.txt
# and extra_questions.txt
# Here, silence_phones.txt and nonsilence_phones.txt are lists of silence and
# non-silence phones respectively (where silence includes various kinds of
# noise, laugh, cough, filled pauses etc., and nonsilence phones includes the
# "real" phones.)
# In each line of those files is a list of phones, and the phones on each line
# are assumed to correspond to the same "base phone", i.e. they will be
# different stress or tone variations of the same basic phone.
# The file "optional_silence.txt" contains just a single phone (typically SIL)
# which is used for optional silence in the lexicon.
# extra_questions.txt might be empty; typically will consist of lists of phones,
# all members of each list with the same stress or tone; and also possibly a
# list for the silence phones.  This will augment the automatically generated
# questions (note: the automatically generated ones will treat all the
# stress/tone versions of a phone the same, so will not "get to ask" about
# stress or tone).
#

# This script adds word-position-dependent phones and constructs a host of other
# derived files, that go in data/lang_subword/.

# Currently it only support the most basic functions.
# Begin configuration section.
num_sil_states=5
num_nonsil_states=3
position_dependent_phones=true
# position_dependent_phones is false also when position dependent phones and word_boundary.txt
# have been generated by another source
share_silence_phones=false  # if true, then share pdfs of different silence
                            # phones together.
sil_prob=0.5
num_extra_phone_disambig_syms=1 # Standard one phone disambiguation symbol is used for optional silence.
                                # Increasing this number does not harm, but is only useful if you later
                                # want to introduce this labels to L_disambig.fst
separator="@@"   # Separator is a suffix or prefix of subword indicating the position of this subword in word.
                 # By default, subword which is not at the end of word would have separator as suffix.
                 # For example: international -> inter@@ nation@@ al

# end configuration sections

echo "$0 $@"  # Print the command line for logging

. utils/parse_options.sh

if [ $# -ne 4 ]; then
  echo "Usage: utils/prepare_lang.sh <dict-src-dir> <oov-dict-entry> <tmp-dir> <lang-dir>"
  echo "e.g.: utils/prepare_lang.sh data/local/dict <SPOKEN_NOISE> data/local/lang data/lang"
  echo "<dict-src-dir> should contain the following files:"
  echo " extra_questions.txt  lexicon.txt nonsilence_phones.txt  optional_silence.txt  silence_phones.txt"
  echo "See http://kaldi-asr.org/doc/data_prep.html#data_prep_lang_creating for more info."
  echo "options: "
  echo "<dict-src-dir> may also, for the grammar-decoding case (see http://kaldi-asr.org/doc/grammar.html)"
  echo "contain a file nonterminals.txt containing symbols like #nonterm:contact_list, one per line."
  echo "     --num-sil-states <number of states>             # default: 5, #states in silence models."
  echo "     --num-nonsil-states <number of states>          # default: 3, #states in non-silence models."
  echo "     --position-dependent-phones (true|false)        # default: true; if true, use _B, _E, _S & _I"
  echo "                                                     # markers on phones to indicate word-internal positions. "
  echo "     --share-silence-phones (true|false)             # default: false; if true, share pdfs of "
  echo "                                                     # all silence phones. "
  echo "     --sil-prob <probability of silence>             # default: 0.5 [must have 0 <= silprob < 1]"
  echo "     --separator <separator>                         # default: @@"
  exit 1;
fi

srcdir=$1
oov_word=$2
tmpdir=$3
dir=$4
mkdir -p $dir $tmpdir $dir/phones

silprob=false
[ -f $srcdir/lexiconp_silprob.txt ] && echo "$0: Currently we do not support word-dependent silence probability." && exit 1;

if [ -f $srcdir/nonterminals.txt ]; then
  echo "$0: Currently we do not support nonterminals" && exit 1;
else
  grammar_opts=
fi

[ -f path.sh ] && . ./path.sh

# Validate dict directory
! utils/validate_dict_dir.pl $srcdir && \
  echo "*Error validating directory $srcdir*" && exit 1;

if [[ ! -f $srcdir/lexicon.txt ]]; then
  echo "**Creating $srcdir/lexicon.txt from $srcdir/lexiconp.txt"
  perl -ape 's/(\S+\s+)\S+\s+(.+)/$1$2/;' < $srcdir/lexiconp.txt > $srcdir/lexicon.txt || exit 1;
fi
if [[ ! -f $srcdir/lexiconp.txt ]]; then
  echo "**Creating $srcdir/lexiconp.txt from $srcdir/lexicon.txt"
  perl -ape 's/(\S+\s+)(.+)/${1}1.0\t$2/;' < $srcdir/lexicon.txt > $srcdir/lexiconp.txt || exit 1;
fi

# Currently The lexicon in dict directory have to be a subword lexicon.
# If the lexicon is for word and is not phonemic, we can not get a subword lexicon without knowing the alignment.
! grep -q $separator $srcdir/lexiconp.txt && \
echo "$0: Warning, this lexicon contains no separator \"$separator\" and may not be a subword lexicon." && exit 1;

# Write the separator into file for future use.
echo $separator > $dir/subword_separator.txt

if ! utils/validate_dict_dir.pl $srcdir >&/dev/null; then
  utils/validate_dict_dir.pl $srcdir  # show the output.
  echo "Validation failed (second time)"
  exit 1;
fi

if $position_dependent_phones; then
  # Create $tmpdir/lexiconp.txt from $srcdir/lexiconp.txt (or
  # $tmpdir/lexiconp_silprob.txt from $srcdir/lexiconp_silprob.txt) by
  # adding the markers _B, _E, _S, _I depending on word position.
  # In this recipe, these markers apply to silence also.
  # Do this starting from lexiconp.txt only.
  if "$silprob"; then
    echo "$0: Currently we do not support word-dependent silence probability" && exit 1;
  else
    utils/lang/make_position_dependent_subword_lexicon.py $srcdir/lexiconp.txt > $tmpdir/lexiconp.txt || exit 1;
  fi

  # create $tmpdir/phone_map.txt
  # this has the format (on each line)
  # <original phone> <version 1 of original phone> <version 2> ...
  # where the versions depend on the position of the phone within a word.
  # For instance, we'd have:
  # AA AA_B AA_E AA_I AA_S
  # for (B)egin, (E)nd, (I)nternal and (S)ingleton
  # and in the case of silence
  # SIL SIL SIL_B SIL_E SIL_I SIL_S
  # [because SIL on its own is one of the variants; this is for when it doesn't
  #  occur inside a word but as an option in the lexicon.]

  # This phone map expands the phone lists into all the word-position-dependent
  # versions of the phone lists.
  cat <(set -f; for x in `cat $srcdir/silence_phones.txt`; do for y in "" "" "_B" "_E" "_I" "_S"; do echo -n "$x$y "; done; echo; done) \
    <(set -f; for x in `cat $srcdir/nonsilence_phones.txt`; do for y in "" "_B" "_E" "_I" "_S"; do echo -n "$x$y "; done; echo; done) \
    > $tmpdir/phone_map.txt
else
  if "$silprob"; then
    echo "$0: Currently we do not support word-dependent silence probability" && exit 1;
  else
    cp $srcdir/lexiconp.txt $tmpdir/lexiconp.txt
  fi

  cat $srcdir/silence_phones.txt $srcdir/nonsilence_phones.txt | \
    awk '{for(n=1;n<=NF;n++) print $n; }' > $tmpdir/phones
  paste -d' ' $tmpdir/phones $tmpdir/phones > $tmpdir/phone_map.txt
fi

mkdir -p $dir/phones  # various sets of phones...

# Sets of phones for use in clustering, and making monophone systems.

if $share_silence_phones; then
  # build a roots file that will force all the silence phones to share the
  # same pdf's. [three distinct states, only the transitions will differ.]
  # 'shared'/'not-shared' means, do we share the 3 states of the HMM
  # in the same tree-root?
  # Sharing across models(phones) is achieved by writing several phones
  # into one line of roots.txt (shared/not-shared doesn't affect this).
  # 'not-shared not-split' means we have separate tree roots for the 3 states,
  # but we never split the tree so they remain stumps,
  # so all phones in the line correspond to the same model.

  cat $srcdir/silence_phones.txt | awk '{printf("%s ", $0); } END{printf("\n");}' | cat - $srcdir/nonsilence_phones.txt | \
    utils/apply_map.pl $tmpdir/phone_map.txt > $dir/phones/sets.txt
  cat $dir/phones/sets.txt | \
    awk '{if(NR==1) print "not-shared", "not-split", $0; else print "shared", "split", $0;}' > $dir/phones/roots.txt
else
  # different silence phones will have different GMMs.  [note: here, all "shared split" means
  # is that we may have one GMM for all the states, or we can split on states.  because they're
  # context-independent phones, they don't see the context.]
  cat $srcdir/{,non}silence_phones.txt | utils/apply_map.pl $tmpdir/phone_map.txt > $dir/phones/sets.txt
  cat $dir/phones/sets.txt | awk '{print "shared", "split", $0;}' > $dir/phones/roots.txt
fi

cat $srcdir/silence_phones.txt | utils/apply_map.pl $tmpdir/phone_map.txt | \
  awk '{for(n=1;n<=NF;n++) print $n;}' > $dir/phones/silence.txt
cat $srcdir/nonsilence_phones.txt | utils/apply_map.pl $tmpdir/phone_map.txt | \
  awk '{for(n=1;n<=NF;n++) print $n;}' > $dir/phones/nonsilence.txt
cp $srcdir/optional_silence.txt $dir/phones/optional_silence.txt
cp $dir/phones/silence.txt $dir/phones/context_indep.txt

# if extra_questions.txt is empty, it's OK.
cat $srcdir/extra_questions.txt 2>/dev/null | utils/apply_map.pl $tmpdir/phone_map.txt \
  >$dir/phones/extra_questions.txt

# Want extra questions about the word-start/word-end stuff. Make it separate for
# silence and non-silence. Probably doesn't matter, as silence will rarely
# be inside a word.
if $position_dependent_phones; then
  for suffix in _B _E _I _S; do
    (set -f; for x in `cat $srcdir/nonsilence_phones.txt`; do echo -n "$x$suffix "; done; echo) >>$dir/phones/extra_questions.txt
  done
  for suffix in "" _B _E _I _S; do
    (set -f; for x in `cat $srcdir/silence_phones.txt`; do echo -n "$x$suffix "; done; echo) >>$dir/phones/extra_questions.txt
  done
fi

# add_lex_disambig.pl is responsible for adding disambiguation symbols to
# the lexicon, for telling us how many disambiguation symbols it used,
# and and also for modifying the unknown-word's pronunciation (if the
# --unk-fst was provided) to the sequence "#1 #2 #3", and reserving those
# disambig symbols for that purpose.
# The #2 will later be replaced with the actual unk model.  The reason
# for the #1 and the #3 is for disambiguation and also to keep the
# FST compact.  If we didn't have the #1, we might have a different copy of
# the unk-model FST, or at least some of its arcs, for each start-state from
# which an <unk> transition comes (instead of per end-state, which is more compact);
# and adding the #3 prevents us from potentially having 2 copies of the unk-model
# FST due to the optional-silence [the last phone of any word gets 2 arcs].

if "$silprob"; then
  echo "$0: Currently we do not support word-dependent silence probability" && exit 1;
else
  ndisambig=$(utils/add_lex_disambig.pl $unk_opt --pron-probs $tmpdir/lexiconp.txt $tmpdir/lexiconp_disambig.txt)
fi
ndisambig=$[$ndisambig+$num_extra_phone_disambig_syms]; # add (at least) one disambig symbol for silence in lexicon FST.
echo $ndisambig > $tmpdir/lex_ndisambig

# Format of lexiconp_disambig.txt:
# !SIL	1.0   SIL_S
# <SPOKEN_NOISE>	1.0   SPN_S #1
# <UNK>	1.0  SPN_S #2
# <NOISE>	1.0  NSN_S
# !EXCLAMATION-POINT	1.0  EH2_B K_I S_I K_I L_I AH0_I M_I EY1_I SH_I AH0_I N_I P_I OY2_I N_I T_E

( for n in `seq 0 $ndisambig`; do echo '#'$n; done ) >$dir/phones/disambig.txt

# Create phone symbol table.
echo "<eps>" | cat - $dir/phones/{silence,nonsilence,disambig}.txt | \
  awk '{n=NR-1; print $1, n;}' > $dir/phones.txt

# Create a file that describes the word-boundary information for
# each phone.  5 categories.
if $position_dependent_phones; then
  cat $dir/phones/{silence,nonsilence}.txt | \
    awk '/_I$/{print $1, "internal"; next;} /_B$/{print $1, "begin"; next; }
         /_S$/{print $1, "singleton"; next;} /_E$/{print $1, "end"; next; }
         {print $1, "nonword";} ' > $dir/phones/word_boundary_moved.txt
else
  # word_boundary.txt might have been generated by another source
  [ -f $srcdir/word_boundary.txt ] && cp $srcdir/word_boundary.txt $dir/phones/word_boundary_moved.txt
fi

# Create word symbol table.
# <s> and </s> are only needed due to the need to rescore lattices with
# ConstArpaLm format language model. They do not normally appear in G.fst or
# L.fst.

if "$silprob"; then
  echo "$0: Currently we do not support word-dependent silence probability" && exit 1;
fi

cat $tmpdir/lexiconp.txt | awk '{print $1}' | sort | uniq  | awk '
  BEGIN {
    print "<eps> 0";
  }
  {
    if ($1 == "<s>") {
      print "<s> is in the vocabulary!" | "cat 1>&2"
      exit 1;
    }
    if ($1 == "</s>") {
      print "</s> is in the vocabulary!" | "cat 1>&2"
      exit 1;
    }
    printf("%s %d\n", $1, NR);
  }
  END {
    printf("#0 %d\n", NR+1);
    printf("<s> %d\n", NR+2);
    printf("</s> %d\n", NR+3);
  }' > $dir/words.txt || exit 1;

# In case there are extra word-level disambiguation symbols they also
# need to be added to words.txt

# format of $dir/words.txt:
# <eps> 0
# a 1
# aa 2
# aarvark 3
# ...

silphone=`cat $srcdir/optional_silence.txt` || exit 1;
[ -z "$silphone" ] && \
  ( echo "You have no optional-silence phone; it is required in the current scripts"
    echo "but you may use the option --sil-prob 0.0 to stop it being used." ) && \
   exit 1;

# create $dir/phones/align_lexicon.{txt,int}.
# This is the method we use for lattice word alignment if we are not
# using word-position-dependent phones.

# First remove pron-probs from the lexicon.
perl -ape 's/(\S+\s+)\S+\s+(.+)/$1$2/;' <$tmpdir/lexiconp.txt >$tmpdir/align_lexicon.txt

# Note: here, $silphone will have no suffix e.g. _S because it occurs as optional-silence,
# and is not part of a word.
[ ! -z "$silphone" ] && echo "<eps> $silphone" >> $tmpdir/align_lexicon.txt

cat $tmpdir/align_lexicon.txt | \
  perl -ane '@A = split; print $A[0], " ", join(" ", @A), "\n";' | sort | uniq > $dir/phones/align_lexicon.txt

# create phones/align_lexicon.int from phones/align_lexicon.txt
cat $dir/phones/align_lexicon.txt | utils/sym2int.pl -f 3- $dir/phones.txt | \
  utils/sym2int.pl -f 1-2 $dir/words.txt > $dir/phones/align_lexicon.int

# Create the basic L.fst without disambiguation symbols, for use
# in training.

if $silprob; then
#  # Add silence probabilities (models the prob. of silence before and after each
#  # word).  On some setups this helps a bit.  See utils/dict_dir_add_pronprobs.sh
#  # and where it's called in the example scripts (run.sh).
  echo "$0: Currently we do not support word-dependnet silence probability" && exit 1;
else
  utils/lang/make_subword_lexicon_fst.py $grammar_opts --sil-prob=$sil_prob --sil-phone=$silphone --position-dependent\
            --separator=$separator $tmpdir/lexiconp.txt | \
    fstcompile --isymbols=$dir/phones.txt --osymbols=$dir/words.txt \
      --keep_isymbols=false --keep_osymbols=false | \
    fstarcsort --sort_type=olabel > $dir/L.fst || exit 1;
fi

# The file oov.txt contains a word that we will map any OOVs to during
# training.
echo "$oov_word" > $dir/oov.txt || exit 1;
cat $dir/oov.txt | utils/sym2int.pl $dir/words.txt >$dir/oov.int || exit 1;
# integer version of oov symbol, used in some scripts.

# the file wdisambig.txt contains a (line-by-line) list of the text-form of the
# disambiguation symbols that are used in the grammar and passed through by the
# lexicon.  At this stage it's hardcoded as '#0', but we're laying the groundwork
# for more generality (which probably would be added by another script).
# wdisambig_words.int contains the corresponding list interpreted by the
# symbol table words.txt, and wdisambig_phones.int contains the corresponding
# list interpreted by the symbol table phones.txt.
echo '#0' >$dir/phones/wdisambig.txt

utils/sym2int.pl $dir/phones.txt <$dir/phones/wdisambig.txt >$dir/phones/wdisambig_phones.int
utils/sym2int.pl $dir/words.txt <$dir/phones/wdisambig.txt >$dir/phones/wdisambig_words.int

# Create these lists of phones in colon-separated integer list form too,
# for purposes of being given to programs as command-line options.
for f in silence nonsilence optional_silence disambig context_indep; do
  utils/sym2int.pl $dir/phones.txt <$dir/phones/$f.txt >$dir/phones/$f.int
  utils/sym2int.pl $dir/phones.txt <$dir/phones/$f.txt | \
   awk '{printf(":%d", $1);} END{printf "\n"}' | sed s/:// > $dir/phones/$f.csl || exit 1;
done

for x in sets extra_questions; do
  utils/sym2int.pl $dir/phones.txt <$dir/phones/$x.txt > $dir/phones/$x.int || exit 1;
done

utils/sym2int.pl -f 3- $dir/phones.txt <$dir/phones/roots.txt \
   > $dir/phones/roots.int || exit 1;

if [ -f $dir/phones/word_boundary_moved.txt ]; then
  utils/sym2int.pl -f 1 $dir/phones.txt <$dir/phones/word_boundary_moved.txt \
    > $dir/phones/word_boundary_moved.int || exit 1;
fi

silphonelist=`cat $dir/phones/silence.csl`
nonsilphonelist=`cat $dir/phones/nonsilence.csl`

# Note: it's OK, after generating the 'lang' directory, to overwrite the topo file
# with another one of your choice if the 'topo' file you want can't be generated by
# utils/gen_topo.pl.  We do this in the 'chain' recipes.  Of course, the 'topo' file
# should cover all the phones.  Try running utils/validate_lang.pl to check that
# everything is OK after modifying the topo file.
utils/gen_topo.pl $num_nonsil_states $num_sil_states $nonsilphonelist $silphonelist >$dir/topo

# Create the lexicon FST with disambiguation symbols, and put it in lang_test.
# There is an extra step where we create a loop to "pass through" the
# disambiguation symbols from G.fst.

if $silprob; then
  echo "$0: Currently we do not support word-dependnet silence probability" && exit 1;
else
  utils/lang/make_subword_lexicon_fst.py $grammar_opts \
       --sil-prob=$sil_prob --sil-phone=$silphone --sil-disambig='#'$ndisambig --position-dependent \
       --separator=$separator $tmpdir/lexiconp_disambig.txt | \
     fstcompile --isymbols=$dir/phones.txt --osymbols=$dir/words.txt \
       --keep_isymbols=false --keep_osymbols=false |   \
     fstaddselfloops  $dir/phones/wdisambig_phones.int $dir/phones/wdisambig_words.int | \
     fstarcsort --sort_type=olabel > $dir/L_disambig.fst || exit 1;
fi

echo "$(basename $0): validating output directory"
! utils/validate_lang.pl $dir && echo "$(basename $0): error validating output" &&  exit 1;

exit 0;