[egs] Add example for Yomdle Farsi OCR (kaldi-asr#2702)

egs/cifar/v1/image/ocr/make_features.py

@@ -43,6 +43,8 @@
 parser.add_argument('--padding', type=int, default=5,
                     help='Number of white pixels to pad on the left'
                     'and right side of the image.')
+parser.add_argument('--num-channels', type=int, default=1,
+                    help='Number of color channels')
 parser.add_argument('--fliplr', type=lambda x: (str(x).lower()=='true'), default=False,
                    help="Flip the image left-right for right to left languages")
 parser.add_argument("--augment", type=lambda x: (str(x).lower()=='true'), default=False,
@@ -84,9 +86,9 @@ def horizontal_pad(im, allowed_lengths = None):
         left_padding = int(padding // 2)
         right_padding = padding - left_padding
     dim_y = im.shape[0] # height
-    im_pad = np.concatenate((255 * np.ones((dim_y, left_padding),
+    im_pad = np.concatenate((255 * np.ones((dim_y, left_padding, args.num_channels),
                                            dtype=int), im), axis=1)
-    im_pad1 = np.concatenate((im_pad, 255 * np.ones((dim_y, right_padding),
+    im_pad1 = np.concatenate((im_pad, 255 * np.ones((dim_y, right_padding, args.num_channels),
                                                     dtype=int)), axis=1)
     return im_pad1
 
@@ -150,7 +152,13 @@ def get_scaled_image_aug(im, mode='normal'):
         if im_horizontal_padded is None:
             num_fail += 1
             continue
-        data = np.transpose(im_horizontal_padded, (1, 0))
+        if args.num_channels == 1:
+            data = np.transpose(im_horizontal_padded, (1, 0))
+        elif args.num_channels == 3:
+            H = im_horizontal_padded.shape[0]
+            W = im_horizontal_padded.shape[1]
+            C = im_horizontal_padded.shape[2]
+            data = np.reshape(np.transpose(im_horizontal_padded, (1, 0, 2)), (W, H * C))
         data = np.divide(data, 255.0)
         num_ok += 1
         write_kaldi_matrix(out_fh, data, image_id)

egs/yomdle_fa/README.txt

@@ -0,0 +1,3 @@
+This directory contains example scripts for OCR on the Yomdle and Slam datasets.
+Training is done on the Yomdle dataset and testing is done on Slam.
+LM rescoring is also done with extra corpus data obtained from various newswires (e.g. Hamshahri)

egs/yomdle_fa/v1/cmd.sh

@@ -0,0 +1,13 @@
+# you can change cmd.sh depending on what type of queue you are using.
+# If you have no queueing system and want to run on a local machine, you
+# can change all instances 'queue.pl' to run.pl (but be careful and run
+# commands one by one: most recipes will exhaust the memory on your
+# machine).  queue.pl works with GridEngine (qsub).  slurm.pl works
+# with slurm.  Different queues are configured differently, with different
+# queue names and different ways of specifying things like memory;
+# to account for these differences you can create and edit the file
+# conf/queue.conf to match your queue's configuration.  Search for
+# conf/queue.conf in http://kaldi-asr.org/doc/queue.html for more information,
+# or search for the string 'default_config' in utils/queue.pl or utils/slurm.pl.
+
+export cmd="queue.pl"

egs/yomdle_fa/v1/image

@@ -0,0 +1 @@
+../../cifar/v1/image/

egs/yomdle_fa/v1/local/augment_data.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+# Copyright   2018 Hossein Hadian
+#             2018 Ashish Arora
+
+# Apache 2.0
+# This script performs data augmentation.
+
+nj=4
+cmd=run.pl
+feat_dim=40
+fliplr=false
+echo "$0 $@"
+
+. ./cmd.sh
+. ./path.sh
+. ./utils/parse_options.sh || exit 1;
+
+srcdir=$1
+outdir=$2
+datadir=$3
+
+mkdir -p $datadir/augmentations
+echo "copying $srcdir to $datadir/augmentations/aug1, allowed length, creating feats.scp"
+
+for set in aug1; do
+  image/copy_data_dir.sh --spk-prefix $set- --utt-prefix $set- \
+    $srcdir $datadir/augmentations/$set
+  cat $srcdir/allowed_lengths.txt > $datadir/augmentations/$set/allowed_lengths.txt
+  local/extract_features.sh --nj $nj --cmd "$cmd" --feat-dim $feat_dim \
+    --fliplr $fliplr --augment true $datadir/augmentations/$set
+done
+
+echo " combine original data and data from different augmentations"
+utils/combine_data.sh --extra-files images.scp $outdir $srcdir $datadir/augmentations/aug1
+cat $srcdir/allowed_lengths.txt > $outdir/allowed_lengths.txt

egs/yomdle_fa/v1/local/bidi.py

@@ -0,0 +1,58 @@
+#!/usr/bin/env python3
+# Copyright   2018 Chun-Chieh Chang
+
+# This script is largely written by Stephen Rawls
+# and uses the python package https://pypi.org/project/PyICU_BiDi/
+# The code leaves right to left text alone and reverses left to right text.
+
+import icu_bidi
+import io
+import sys
+import unicodedata
+# R=strong right-to-left;  AL=strong arabic right-to-left
+rtl_set =  set(chr(i) for i in range(sys.maxunicode)
+               if unicodedata.bidirectional(chr(i)) in ['R','AL'])
+def determine_text_direction(text):
+    # Easy case first
+    for char in text:
+        if char in rtl_set:
+            return icu_bidi.UBiDiLevel.UBIDI_RTL
+    # If we made it here we did not encounter any strongly rtl char
+    return icu_bidi.UBiDiLevel.UBIDI_LTR
+
+def utf8_visual_to_logical(text):
+    text_dir = determine_text_direction(text)
+
+    bidi = icu_bidi.Bidi()
+    bidi.inverse = True
+    bidi.reordering_mode = icu_bidi.UBiDiReorderingMode.UBIDI_REORDER_INVERSE_LIKE_DIRECT
+    bidi.reordering_options = icu_bidi.UBiDiReorderingOption.UBIDI_OPTION_DEFAULT # icu_bidi.UBiDiReorderingOption.UBIDI_OPTION_INSERT_MARKS
+
+    bidi.set_para(text, text_dir, None)
+
+    res = bidi.get_reordered(0 | icu_bidi.UBidiWriteReorderedOpt.UBIDI_DO_MIRRORING | icu_bidi.UBidiWriteReorderedOpt.UBIDI_KEEP_BASE_COMBINING)
+
+    return res
+
+def utf8_logical_to_visual(text):
+    text_dir = determine_text_direction(text)
+
+    bidi = icu_bidi.Bidi()
+
+    bidi.reordering_mode = icu_bidi.UBiDiReorderingMode.UBIDI_REORDER_DEFAULT
+    bidi.reordering_options = icu_bidi.UBiDiReorderingOption.UBIDI_OPTION_DEFAULT  #icu_bidi.UBiDiReorderingOption.UBIDI_OPTION_INSERT_MARKS
+
+    bidi.set_para(text, text_dir, None)
+
+    res = bidi.get_reordered(0 | icu_bidi.UBidiWriteReorderedOpt.UBIDI_DO_MIRRORING | icu_bidi.UBidiWriteReorderedOpt.UBIDI_KEEP_BASE_COMBINING)
+
+    return res
+
+
+##main##
+sys.stdin = io.TextIOWrapper(sys.stdin.buffer, encoding="utf8")
+sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding="utf8")
+for line in sys.stdin:
+    line = line.strip()
+    line = utf8_logical_to_visual(line)[::-1]
+    sys.stdout.write(line + '\n')

egs/yomdle_fa/v1/local/chain/compare_wer.sh

@@ -0,0 +1,67 @@
+#!/bin/bash
+
+# this script is used for comparing decoding results between systems.
+# e.g. local/chain/compare_wer.sh exp/chain/cnn{1a,1b}
+
+# Copyright      2017  Chun Chieh Chang
+#                2017  Ashish Arora
+
+if [ $# == 0 ]; then
+  echo "Usage: $0: <dir1> [<dir2> ... ]"
+  echo "e.g.: $0 exp/chain/cnn{1a,1b}"
+  exit 1
+fi
+
+echo "# $0 $*"
+used_epochs=false
+
+echo -n "# System                     "
+for x in $*; do   printf "% 10s" " $(basename $x)";   done
+echo
+
+echo -n "# WER                        "
+for x in $*; do
+  wer=$(cat $x/decode_test/scoring_kaldi/best_wer | awk '{print $2}')
+  printf "% 10s" $wer
+done
+echo
+
+echo -n "# CER                        "
+for x in $*; do
+  cer=$(cat $x/decode_test/scoring_kaldi/best_cer | awk '{print $2}')
+  printf "% 10s" $cer
+done
+echo
+
+
+if $used_epochs; then
+  exit 0;  # the diagnostics aren't comparable between regular and discriminatively trained systems.
+fi
+
+echo -n "# Final train prob           "
+for x in $*; do
+  prob=$(grep Overall $x/log/compute_prob_train.final.log | grep -v xent | awk '{printf("%.4f", $8)}')
+  printf "% 10s" $prob
+done
+echo
+
+echo -n "# Final valid prob           "
+for x in $*; do
+  prob=$(grep Overall $x/log/compute_prob_valid.final.log | grep -v xent | awk '{printf("%.4f", $8)}')
+  printf "% 10s" $prob
+done
+echo
+
+echo -n "# Final train prob (xent)    "
+for x in $*; do
+  prob=$(grep Overall $x/log/compute_prob_train.final.log | grep -w xent | awk '{printf("%.4f", $8)}')
+  printf "% 10s" $prob
+done
+echo
+
+echo -n "# Final valid prob (xent)    "
+for x in $*; do
+  prob=$(grep Overall $x/log/compute_prob_valid.final.log | grep -w xent | awk '{printf("%.4f", $8)}')
+  printf "% 10s" $prob
+done
+echo