diff --git a/.DS_Store b/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..983053dc788bd10926c6731ab98d714eebd335be
Binary files /dev/null and b/.DS_Store differ
diff --git a/egs/.DS_Store b/egs/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..e04067a2d4553d65b59f96c8f73421ef6bf15192
Binary files /dev/null and b/egs/.DS_Store differ
diff --git a/egs/digits b/egs/digits
new file mode 160000
index 0000000000000000000000000000000000000000..7a670411cf604463bf004beef273467c019d033b
--- /dev/null
+++ b/egs/digits
@@ -0,0 +1 @@
+Subproject commit 7a670411cf604463bf004beef273467c019d033b
diff --git a/egs/kaldi_toy_example b/egs/kaldi_toy_example
new file mode 160000
index 0000000000000000000000000000000000000000..90246e855df9742662efc639e2aa11a042957290
--- /dev/null
+++ b/egs/kaldi_toy_example
@@ -0,0 +1 @@
+Subproject commit 90246e855df9742662efc639e2aa11a042957290
diff --git a/egs/voxceleb/.DS_Store b/egs/voxceleb/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..fd3b239592f6b822947e0fbd2575dd5a1ebf4a24
Binary files /dev/null and b/egs/voxceleb/.DS_Store differ
diff --git a/egs/voxceleb/v1/.DS_Store b/egs/voxceleb/v1/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..afcae070c37829f7a5a1ecb02beec75813e1195d
Binary files /dev/null and b/egs/voxceleb/v1/.DS_Store differ
diff --git a/egs/voxceleb/v2/.DS_Store b/egs/voxceleb/v2/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..16ba07ade7f9a22bb4e381774f0f19aa150ecd0f
Binary files /dev/null and b/egs/voxceleb/v2/.DS_Store differ
diff --git a/egs/voxceleb/v2/local/make_voxceleb1.pl b/egs/voxceleb/v2/local/make_voxceleb1.pl
new file mode 100755
index 0000000000000000000000000000000000000000..2268c20ab52aa62f47d2cb63ec13cc6073ff324d
--- /dev/null
+++ b/egs/voxceleb/v2/local/make_voxceleb1.pl
@@ -0,0 +1,130 @@
+#!/usr/bin/perl
+#
+# Copyright 2018 Ewald Enzinger
+# 2018 David Snyder
+#
+# Usage: make_voxceleb1.pl /export/voxceleb1 data/
+
+if (@ARGV != 2) {
+ print STDERR "Usage: $0 <path-to-voxceleb1> <path-to-data-dir>\n";
+ print STDERR "e.g. $0 /export/voxceleb1 data/\n";
+ exit(1);
+}
+
+($data_base, $out_dir) = @ARGV;
+my $out_test_dir = "$out_dir/voxceleb1_test";
+my $out_train_dir = "$out_dir/voxceleb1_train";
+
+if (system("mkdir -p $out_test_dir") != 0) {
+ die "Error making directory $out_test_dir";
+}
+
+if (system("mkdir -p $out_train_dir") != 0) {
+ die "Error making directory $out_train_dir";
+}
+
+opendir my $dh, "$data_base/voxceleb1_wav" or die "Cannot open directory: $!";
+my @spkr_dirs = grep {-d "$data_base/voxceleb1_wav/$_" && ! /^\.{1,2}$/} readdir($dh);
+closedir $dh;
+
+if (! -e "$data_base/voxceleb1_test.txt") {
+ system("wget -O $data_base/voxceleb1_test.txt http://www.openslr.org/resources/49/voxceleb1_test.txt");
+}
+
+if (! -e "$data_base/vox1_meta.csv") {
+ system("wget -O $data_base/vox1_meta.csv http://www.openslr.org/resources/49/vox1_meta.csv");
+}
+
+open(TRIAL_IN, "<", "$data_base/voxceleb1_test.txt") or die "Could not open the verification trials file $data_base/voxceleb1_test.txt";
+open(META_IN, "<", "$data_base/vox1_meta.csv") or die "Could not open the meta data file $data_base/vox1_meta.csv";
+open(SPKR_TEST, ">", "$out_test_dir/utt2spk") or die "Could not open the output file $out_test_dir/utt2spk";
+open(WAV_TEST, ">", "$out_test_dir/wav.scp") or die "Could not open the output file $out_test_dir/wav.scp";
+open(SPKR_TRAIN, ">", "$out_train_dir/utt2spk") or die "Could not open the output file $out_train_dir/utt2spk";
+open(WAV_TRAIN, ">", "$out_train_dir/wav.scp") or die "Could not open the output file $out_train_dir/wav.scp";
+open(TRIAL_OUT, ">", "$out_test_dir/trials") or die "Could not open the output file $out_test_dir/trials";
+
+my %id2spkr = ();
+while (<META_IN>) {
+ chomp;
+ my ($vox_id, $spkr_id, $gender, $nation, $set) = split;
+ $id2spkr{$vox_id} = $spkr_id;
+}
+
+my $test_spkrs = ();
+while (<TRIAL_IN>) {
+ chomp;
+ my ($tar_or_non, $path1, $path2) = split;
+
+ # Create entry for left-hand side of trial
+ my ($spkr_id, $filename) = split('/', $path1);
+ my $rec_id = substr($filename, 0, 11);
+ my $segment = substr($filename, 12, 7);
+ my $utt_id1 = "$spkr_id-$rec_id-$segment";
+ $test_spkrs{$spkr_id} = ();
+
+ # Create entry for right-hand side of trial
+ my ($spkr_id, $filename) = split('/', $path2);
+ my $rec_id = substr($filename, 0, 11);
+ my $segment = substr($filename, 12, 7);
+ my $utt_id2 = "$spkr_id-$rec_id-$segment";
+ $test_spkrs{$spkr_id} = ();
+
+ my $target = "nontarget";
+ if ($tar_or_non eq "1") {
+ $target = "target";
+ }
+ print TRIAL_OUT "$utt_id1 $utt_id2 $target\n";
+}
+
+foreach (@spkr_dirs) {
+ my $spkr_id = $_;
+ my $new_spkr_id = $spkr_id;
+ # If we're using a newer version of VoxCeleb1, we need to "deanonymize"
+ # the speaker labels.
+ if (exists $id2spkr{$spkr_id}) {
+ $new_spkr_id = $id2spkr{$spkr_id};
+ }
+ opendir my $dh, "$data_base/voxceleb1_wav/$spkr_id/" or die "Cannot open directory: $!";
+ my @files = map{s/\.[^.]+$//;$_}grep {/\.wav$/} readdir($dh);
+ closedir $dh;
+ foreach (@files) {
+ my $filename = $_;
+ my $rec_id = substr($filename, 0, 11);
+ my $segment = substr($filename, 12, 7);
+ my $wav = "$data_base/voxceleb1_wav/$spkr_id/$filename.wav";
+ my $utt_id = "$new_spkr_id-$rec_id-$segment";
+ if (exists $test_spkrs{$new_spkr_id}) {
+ print WAV_TEST "$utt_id", " $wav", "\n";
+ print SPKR_TEST "$utt_id", " $new_spkr_id", "\n";
+ } else {
+ print WAV_TRAIN "$utt_id", " $wav", "\n";
+ print SPKR_TRAIN "$utt_id", " $new_spkr_id", "\n";
+ }
+ }
+}
+
+close(SPKR_TEST) or die;
+close(WAV_TEST) or die;
+close(SPKR_TRAIN) or die;
+close(WAV_TRAIN) or die;
+close(TRIAL_OUT) or die;
+close(TRIAL_IN) or die;
+close(META_IN) or die;
+
+if (system(
+ "utils/utt2spk_to_spk2utt.pl $out_test_dir/utt2spk >$out_test_dir/spk2utt") != 0) {
+ die "Error creating spk2utt file in directory $out_test_dir";
+}
+system("env LC_COLLATE=C utils/fix_data_dir.sh $out_test_dir");
+if (system("env LC_COLLATE=C utils/validate_data_dir.sh --no-text --no-feats $out_test_dir") != 0) {
+ die "Error validating directory $out_test_dir";
+}
+
+if (system(
+ "utils/utt2spk_to_spk2utt.pl $out_train_dir/utt2spk >$out_train_dir/spk2utt") != 0) {
+ die "Error creating spk2utt file in directory $out_train_dir";
+}
+system("env LC_COLLATE=C utils/fix_data_dir.sh $out_train_dir");
+if (system("env LC_COLLATE=C utils/validate_data_dir.sh --no-text --no-feats $out_train_dir") != 0) {
+ die "Error validating directory $out_train_dir";
+}
diff --git a/egs/voxceleb/v2/local/make_voxceleb1_v2.pl b/egs/voxceleb/v2/local/make_voxceleb1_v2.pl
new file mode 100755
index 0000000000000000000000000000000000000000..221507870be6e6edc15c3b0a7cf4569fa5fd472b
--- /dev/null
+++ b/egs/voxceleb/v2/local/make_voxceleb1_v2.pl
@@ -0,0 +1,125 @@
+#!/usr/bin/perl
+#
+# Copyright 2018 Ewald Enzinger
+# 2018 David Snyder
+# 2019 Soonshin Seo
+#
+# Usage: make_voxceleb1_v2.pl /export/voxceleb1 dev data/dev
+#
+# The VoxCeleb1 corpus underwent several updates that changed the directory and speaker ID format.
+# The script 'make_voxceleb1.pl' works for the oldest version of the corpus.
+# This script should be used if you've downloaded the corpus recently.
+
+if (@ARGV != 3) {
+ print STDERR "Usage: $0 <path-to-voxceleb1> <dataset> <path-to-data-dir>\n";
+ print STDERR "e.g. $0 /export/voxceleb1 dev data/dev\n";
+ exit(1);
+}
+
+($data_base, $dataset, $out_dir) = @ARGV;
+
+if ("$dataset" ne "dev" && "$dataset" ne "test") {
+ die "dataset parameter must be 'dev' or 'test'!";
+}
+
+if (system("mkdir -p $out_dir") != 0) {
+ die "Error making directory $out_dir";
+}
+
+opendir my $dh, "$data_base/$dataset/wav" or die "Cannot open directory: $!";
+my @spkr_dirs = grep {-d "$data_base/$dataset/wav/$_" && ! /^\.{1,2}$/} readdir($dh);
+closedir $dh;
+
+if ($dataset eq "dev"){
+ open(SPKR_TRAIN, ">", "$out_dir/utt2spk") or die "could not open the output file $out_dir/utt2spk";
+ open(WAV_TRAIN, ">", "$out_dir/wav.scp") or die "could not open the output file $out_dir/wav.scp";
+
+ foreach (@spkr_dirs) {
+ my $spkr_id = $_;
+ opendir my $dh, "$data_base/$dataset/wav/$spkr_id/" or die "Cannot open directory: $!";
+ my @rec_dirs = grep {-d "$data_base/$dataset/wav/$spkr_id/$_" && ! /^\.{1,2}$/} readdir($dh);
+ closedir $dh;
+ foreach (@rec_dirs) {
+ my $rec_id = $_;
+ opendir my $dh, "$data_base/$dataset/wav/$spkr_id/$rec_id/" or die "Cannot open directory: $!";
+ my @files = map{s/\.[^.]+$//;$_}grep {/\.wav$/} readdir($dh);
+ closedir $dh;
+ foreach (@files) {
+ my $name = $_;
+ my $wav = "$data_base/$dataset/wav/$spkr_id/$rec_id/$name.wav";
+ my $utt_id = "$spkr_id-$rec_id-$name";
+ print WAV_TRAIN "$utt_id", " $wav", "\n";
+ print SPKR_TRAIN "$utt_id", " $spkr_id", "\n";
+ }
+ }
+ }
+ close(SPKR_TRAIN) or die;
+ close(WAV_TRAIN) or die;
+}
+
+if ($dataset eq "test"){
+ if (! -e "$data_base/voxceleb1_test_v2.txt") {
+ system("wget -O $data_base/voxceleb1_test_v2.txt http://www.openslr.org/resources/49/voxceleb1_test_v2.txt");
+ }
+
+ open(TRIAL_IN, "<", "$data_base/voxceleb1_test_v2.txt") or die "could not open the verification trials file $data_base/voxceleb1_test_v2.txt";
+ open(TRIAL_OUT, ">", "$out_dir/trials") or die "Could not open the output file $out_test_dir/trials";
+ open(SPKR_TEST, ">", "$out_dir/utt2spk") or die "could not open the output file $out_dir/utt2spk";
+ open(WAV_TEST, ">", "$out_dir/wav.scp") or die "could not open the output file $out_dir/wav.scp";
+
+ my $test_spkrs = ();
+ while (<TRIAL_IN>) {
+ chomp;
+ my ($tar_or_non, $path1, $path2) = split;
+ # Create entry for left-hand side of trial
+ my ($spkr_id, $rec_id, $name) = split('/', $path1);
+ $name =~ s/\.wav$//g;
+ my $utt_id1 = "$spkr_id-$rec_id-$name";
+ $test_spkrs{$spkr_id} = ();
+
+ # Create entry for right-hand side of trial
+ my ($spkr_id, $rec_id, $name) = split('/', $path2);
+ $name =~ s/\.wav$//g;
+ my $utt_id2 = "$spkr_id-$rec_id-$name";
+ $test_spkrs{$spkr_id} = ();
+
+ my $target = "nontarget";
+ if ($tar_or_non eq "1") {
+ $target = "target";
+ }
+ print TRIAL_OUT "$utt_id1 $utt_id2 $target\n";
+ }
+
+ foreach (@spkr_dirs) {
+ my $spkr_id = $_;
+ opendir my $dh, "$data_base/$dataset/wav/$spkr_id/" or die "Cannot open directory: $!";
+ my @rec_dirs = grep {-d "$data_base/$dataset/wav/$spkr_id/$_" && ! /^\.{1,2}$/} readdir($dh);
+ closedir $dh;
+ foreach (@rec_dirs) {
+ my $rec_id = $_;
+ opendir my $dh, "$data_base/$dataset/wav/$spkr_id/$rec_id/" or die "Cannot open directory: $!";
+ my @files = map{s/\.[^.]+$//;$_}grep {/\.wav$/} readdir($dh);
+ closedir $dh;
+ foreach (@files) {
+ my $name = $_;
+ my $wav = "$data_base/$dataset/wav/$spkr_id/$rec_id/$name.wav";
+ my $utt_id = "$spkr_id-$rec_id-$name";
+ print WAV_TEST "$utt_id", " $wav", "\n";
+ print SPKR_TEST "$utt_id", " $spkr_id", "\n";
+ }
+ }
+ }
+ close(SPKR_TEST) or die;
+ close(WAV_TEST) or die;
+ close(TRIAL_OUT) or die;
+ close(TRIAL_IN) or die;
+}
+
+if (system(
+ "utils/utt2spk_to_spk2utt.pl $out_dir/utt2spk >$out_dir/spk2utt") != 0) {
+ die "Error creating spk2utt file in directory $out_dir";
+}
+system("env LC_COLLATE=C utils/fix_data_dir.sh $out_dir");
+if (system("env LC_COLLATE=C utils/validate_data_dir.sh --no-text --no-feats $out_dir") != 0) {
+ die "Error validating directory $out_dir";
+}
diff --git a/egs/voxceleb/v2/local/make_voxceleb2.pl b/egs/voxceleb/v2/local/make_voxceleb2.pl
new file mode 100755
index 0000000000000000000000000000000000000000..34c1591eba3896e18ec1591443eb55cb429f24ff
--- /dev/null
+++ b/egs/voxceleb/v2/local/make_voxceleb2.pl
@@ -0,0 +1,70 @@
+#!/usr/bin/perl
+#
+# Copyright 2018 Ewald Enzinger
+#
+# Usage: make_voxceleb2.pl /export/voxceleb2 dev data/dev
+#
+# Note: This script requires ffmpeg to be installed and its location included in $PATH.
+
+if (@ARGV != 3) {
+ print STDERR "Usage: $0 <path-to-voxceleb2> <dataset> <path-to-data-dir>\n";
+ print STDERR "e.g. $0 /export/voxceleb2 dev data/dev\n";
+ exit(1);
+}
+
+# Check that ffmpeg is installed.
+if (`which ffmpeg` eq "") {
+ die "Error: this script requires that ffmpeg is installed.";
+}
+
+($data_base, $dataset, $out_dir) = @ARGV;
+
+if ("$dataset" ne "dev" && "$dataset" ne "test") {
+ die "dataset parameter must be 'dev' or 'test'!";
+}
+
+opendir my $dh, "$data_base/$dataset/aac" or die "Cannot open directory: $!";
+my @spkr_dirs = grep {-d "$data_base/$dataset/aac/$_" && ! /^\.{1,2}$/} readdir($dh);
+closedir $dh;
+
+if (system("mkdir -p $out_dir") != 0) {
+ die "Error making directory $out_dir";
+}
+
+open(SPKR, ">", "$out_dir/utt2spk") or die "Could not open the output file $out_dir/utt2spk";
+open(WAV, ">", "$out_dir/wav.scp") or die "Could not open the output file $out_dir/wav.scp";
+
+foreach (@spkr_dirs) {
+ my $spkr_id = $_;
+
+ opendir my $dh, "$data_base/$dataset/aac/$spkr_id/" or die "Cannot open directory: $!";
+ my @rec_dirs = grep {-d "$data_base/$dataset/aac/$spkr_id/$_" && ! /^\.{1,2}$/} readdir($dh);
+ closedir $dh;
+
+ foreach (@rec_dirs) {
+ my $rec_id = $_;
+
+ opendir my $dh, "$data_base/$dataset/aac/$spkr_id/$rec_id/" or die "Cannot open directory: $!";
+ my @files = map{s/\.[^.]+$//;$_}grep {/\.m4a$/} readdir($dh);
+ closedir $dh;
+
+ foreach (@files) {
+ my $name = $_;
+ my $wav = "ffmpeg -v 8 -i $data_base/$dataset/aac/$spkr_id/$rec_id/$name.m4a -f wav -acodec pcm_s16le -|";
+ my $utt_id = "$spkr_id-$rec_id-$name";
+ print WAV "$utt_id", " $wav", "\n";
+ print SPKR "$utt_id", " $spkr_id", "\n";
+ }
+ }
+}
+close(SPKR) or die;
+close(WAV) or die;
+
+if (system(
+ "utils/utt2spk_to_spk2utt.pl $out_dir/utt2spk >$out_dir/spk2utt") != 0) {
+ die "Error creating spk2utt file in directory $out_dir";
+}
+system("env LC_COLLATE=C utils/fix_data_dir.sh $out_dir");
+if (system("env LC_COLLATE=C utils/validate_data_dir.sh --no-text --no-feats $out_dir") != 0) {
+ die "Error validating directory $out_dir";
+}
diff --git a/egs/voxceleb/v2/local/nnet3/xvector/prepare_feats_for_egs.sh b/egs/voxceleb/v2/local/nnet3/xvector/prepare_feats_for_egs.sh
new file mode 100755
index 0000000000000000000000000000000000000000..a899ea7e952f35182dfaf4849e70316f6ae9b959
--- /dev/null
+++ b/egs/voxceleb/v2/local/nnet3/xvector/prepare_feats_for_egs.sh
@@ -0,0 +1,84 @@
+#!/usr/bin/env bash
+#
+# Copied from egs/sre16/v1/local/nnet3/xvector/prepare_feats_for_egs.sh (commit 3ea534070fd2cccd2e4ee21772132230033022ce).
+#
+# Apache 2.0.
+
+# This script applies sliding window cmvn and removes silence frames. This
+# is performed on the raw features prior to generating examples for training
+# the xvector system.
+
+nj=40
+cmd="run.pl"
+stage=0
+norm_vars=false
+center=true
+compress=true
+cmn_window=300
+
+echo "$0 $@" # Print the command line for logging
+
+if [ -f path.sh ]; then . ./path.sh; fi
+. parse_options.sh || exit 1;
+if [ $# != 3 ]; then
+ echo "Usage: $0 <in-data-dir> <out-data-dir> <feat-dir>"
+ echo "e.g.: $0 data/train data/train_no_sil exp/make_xvector_features"
+ echo "Options: "
+ echo " --nj <nj> # number of parallel jobs"
+ echo " --cmd (utils/run.pl|utils/queue.pl <queue opts>) # how to run jobs."
+ echo " --norm-vars <true|false> # If true, normalize variances in the sliding window cmvn"
+ exit 1;
+fi
+
+data_in=$1
+data_out=$2
+dir=$3
+
+name=`basename $data_in`
+
+for f in $data_in/feats.scp $data_in/vad.scp ; do
+ [ ! -f $f ] && echo "$0: No such file $f" && exit 1;
+done
+
+# Set various variables.
+mkdir -p $dir/log
+mkdir -p $data_out
+featdir=$(utils/make_absolute.sh $dir)
+
+if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $featdir/storage ]; then
+ utils/create_split_dir.pl \
+ /export/b{14,15,16,17}/$USER/kaldi-data/egs/voxceleb2/v2/xvector-$(date +'%m_%d_%H_%M')/xvector_feats/storage $featdir/storage
+fi
+
+for n in $(seq $nj); do
+ # the next command does nothing unless $featdir/storage/ exists, see
+ # utils/create_data_link.pl for more info.
+ utils/create_data_link.pl $featdir/xvector_feats_${name}.${n}.ark
+done
+
+cp $data_in/utt2spk $data_out/utt2spk
+cp $data_in/spk2utt $data_out/spk2utt
+cp $data_in/wav.scp $data_out/wav.scp
+
+write_num_frames_opt="--write-num-frames=ark,t:$featdir/log/utt2num_frames.JOB"
+
+sdata_in=$data_in/split$nj;
+utils/split_data.sh $data_in $nj || exit 1;
+
+$cmd JOB=1:$nj $dir/log/create_xvector_feats_${name}.JOB.log \
+ apply-cmvn-sliding --norm-vars=false --center=true --cmn-window=$cmn_window \
+ scp:${sdata_in}/JOB/feats.scp ark:- \| \
+ select-voiced-frames ark:- scp,s,cs:${sdata_in}/JOB/vad.scp ark:- \| \
+ copy-feats --compress=$compress $write_num_frames_opt ark:- \
+ ark,scp:$featdir/xvector_feats_${name}.JOB.ark,$featdir/xvector_feats_${name}.JOB.scp || exit 1;
+
+for n in $(seq $nj); do
+ cat $featdir/xvector_feats_${name}.$n.scp || exit 1;
+done > ${data_out}/feats.scp || exit 1
+
+for n in $(seq $nj); do
+ cat $featdir/log/utt2num_frames.$n || exit 1;
+done > $data_out/utt2num_frames || exit 1
+rm $featdir/log/utt2num_frames.*
+
+echo "$0: Succeeded creating xvector features for $name"
diff --git a/egs/voxceleb/v2/local/nnet3/xvector/run_xvector.sh b/egs/voxceleb/v2/local/nnet3/xvector/run_xvector.sh
new file mode 120000
index 0000000000000000000000000000000000000000..585b63fd2dd8ea0937b929f980df2b0b1b561288
--- /dev/null
+++ b/egs/voxceleb/v2/local/nnet3/xvector/run_xvector.sh
@@ -0,0 +1 @@
+tuning/run_xvector_1a.sh
\ No newline at end of file
diff --git a/egs/voxceleb/v2/local/nnet3/xvector/tuning/run_xvector_1a.sh b/egs/voxceleb/v2/local/nnet3/xvector/tuning/run_xvector_1a.sh
new file mode 100755
index 0000000000000000000000000000000000000000..a7bb0cdd4329c0e4db997889c515d7e72bc7d1cc
--- /dev/null
+++ b/egs/voxceleb/v2/local/nnet3/xvector/tuning/run_xvector_1a.sh
@@ -0,0 +1,155 @@
+#!/usr/bin/env bash
+# Copyright 2017 David Snyder
+# 2017 Johns Hopkins University (Author: Daniel Garcia-Romero)
+# 2017 Johns Hopkins University (Author: Daniel Povey)
+#
+# Copied from egs/sre16/v1/local/nnet3/xvector/tuning/run_xvector_1a.sh (commit e082c17d4a8f8a791428ae4d9f7ceb776aef3f0b).
+#
+# Apache 2.0.
+
+# This script trains a DNN similar to the recipe described in
+# http://www.danielpovey.com/files/2018_icassp_xvectors.pdf
+
+. ./cmd.sh
+set -e
+
+stage=1
+train_stage=0
+use_gpu=true
+remove_egs=false
+
+data=data/train
+nnet_dir=exp/xvector_nnet_1a/
+egs_dir=exp/xvector_nnet_1a/egs
+
+. ./path.sh
+. ./cmd.sh
+. ./utils/parse_options.sh
+
+num_pdfs=$(awk '{print $2}' $data/utt2spk | sort | uniq -c | wc -l)
+
+# Now we create the nnet examples using sid/nnet3/xvector/get_egs.sh.
+# The argument --num-repeats is related to the number of times a speaker
+# repeats per archive. If it seems like you're getting too many archives
+# (e.g., more than 200) try increasing the --frames-per-iter option. The
+# arguments --min-frames-per-chunk and --max-frames-per-chunk specify the
+# minimum and maximum length (in terms of number of frames) of the features
+# in the examples.
+#
+# To make sense of the egs script, it may be necessary to put an "exit 1"
+# command immediately after stage 3. Then, inspect
+# exp/<your-dir>/egs/temp/ranges.* . The ranges files specify the examples that
+# will be created, and which archives they will be stored in. Each line of
+# ranges.* has the following form:
+# <utt-id> <local-ark-indx> <global-ark-indx> <start-frame> <end-frame> <spk-id>
+# For example:
+# 100304-f-sre2006-kacg-A 1 2 4079 881 23
+
+# If you're satisfied with the number of archives (e.g., 50-150 archives is
+# reasonable) and with the number of examples per speaker (e.g., 1000-5000
+# is reasonable) then you can let the script continue to the later stages.
+# Otherwise, try increasing or decreasing the --num-repeats option. You might
+# need to fiddle with --frames-per-iter. Increasing this value decreases the
+# the number of archives and increases the number of examples per archive.
+# Decreasing this value increases the number of archives, while decreasing the
+# number of examples per archive.
+if [ $stage -le 6 ]; then
+ echo "$0: Getting neural network training egs";
+ # dump egs.
+ if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $egs_dir/storage ]; then
+ utils/create_split_dir.pl \
+ /export/b{03,04,05,06}/$USER/kaldi-data/egs/voxceleb2/v2/xvector-$(date +'%m_%d_%H_%M')/$egs_dir/storage $egs_dir/storage
+ fi
+ sid/nnet3/xvector/get_egs.sh --cmd "$train_cmd" \
+ --nj 8 \
+ --stage 0 \
+ --frames-per-iter 1000000000 \
+ --frames-per-iter-diagnostic 100000 \
+ --min-frames-per-chunk 200 \
+ --max-frames-per-chunk 400 \
+ --num-diagnostic-archives 3 \
+ --num-repeats 50 \
+ "$data" $egs_dir
+fi
+
+if [ $stage -le 7 ]; then
+ echo "$0: creating neural net configs using the xconfig parser";
+ num_targets=$(wc -w $egs_dir/pdf2num | awk '{print $1}')
+ feat_dim=$(cat $egs_dir/info/feat_dim)
+
+ # This chunk-size corresponds to the maximum number of frames the
+ # stats layer is able to pool over. In this script, it corresponds
+ # to 100 seconds. If the input recording is greater than 100 seconds,
+ # we will compute multiple xvectors from the same recording and average
+ # to produce the final xvector.
+ max_chunk_size=10000
+
+ # The smallest number of frames we're comfortable computing an xvector from.
+ # Note that the hard minimum is given by the left and right context of the
+ # frame-level layers.
+ min_chunk_size=25
+ mkdir -p $nnet_dir/configs
+ cat <<EOF > $nnet_dir/configs/network.xconfig
+ # please note that it is important to have input layer with the name=input
+
+ # The frame-level layers
+ input dim=${feat_dim} name=input
+ relu-batchnorm-layer name=tdnn1 input=Append(-2,-1,0,1,2) dim=512
+ relu-batchnorm-layer name=tdnn2 input=Append(-2,0,2) dim=512
+ relu-batchnorm-layer name=tdnn3 input=Append(-3,0,3) dim=512
+ relu-batchnorm-layer name=tdnn4 dim=512
+ relu-batchnorm-layer name=tdnn5 dim=1500
+
+ # The stats pooling layer. Layers after this are segment-level.
+ # In the config below, the first and last argument (0, and ${max_chunk_size})
+ # means that we pool over an input segment starting at frame 0
+ # and ending at frame ${max_chunk_size} or earlier. The other arguments (1:1)
+ # mean that no subsampling is performed.
+ stats-layer name=stats config=mean+stddev(0:1:1:${max_chunk_size})
+
+ # This is where we usually extract the embedding (aka xvector) from.
+ relu-batchnorm-layer name=tdnn6 dim=512 input=stats
+
+ # This is where another layer the embedding could be extracted
+ # from, but usually the previous one works better.
+ relu-batchnorm-layer name=tdnn7 dim=512
+ output-layer name=output include-log-softmax=true dim=${num_targets}
+EOF
+
+ steps/nnet3/xconfig_to_configs.py \
+ --xconfig-file $nnet_dir/configs/network.xconfig \
+ --config-dir $nnet_dir/configs/
+ cp $nnet_dir/configs/final.config $nnet_dir/nnet.config
+
+ # These three files will be used by sid/nnet3/xvector/extract_xvectors.sh
+ echo "output-node name=output input=tdnn6.affine" > $nnet_dir/extract.config
+ echo "$max_chunk_size" > $nnet_dir/max_chunk_size
+ echo "$min_chunk_size" > $nnet_dir/min_chunk_size
+fi
+
+dropout_schedule='0,0@0.20,0.1@0.50,0'
+srand=123
+if [ $stage -le 8 ]; then
+ steps/nnet3/train_raw_dnn.py --stage=$train_stage \
+ --cmd="$train_cmd" \
+ --trainer.optimization.proportional-shrink 10 \
+ --trainer.optimization.momentum=0.5 \
+ --trainer.optimization.num-jobs-initial=3 \
+ --trainer.optimization.num-jobs-final=8 \
+ --trainer.optimization.initial-effective-lrate=0.001 \
+ --trainer.optimization.final-effective-lrate=0.0001 \
+ --trainer.optimization.minibatch-size=64 \
+ --trainer.srand=$srand \
+ --trainer.max-param-change=2 \
+ --trainer.num-epochs=3 \
+ --trainer.dropout-schedule="$dropout_schedule" \
+ --trainer.shuffle-buffer-size=1000 \
+ --egs.frames-per-eg=1 \
+ --egs.dir="$egs_dir" \
+ --cleanup.remove-egs $remove_egs \
+ --cleanup.preserve-model-interval=10 \
+ --use-gpu=true \
+ --dir=$nnet_dir || exit 1;
+fi
+
+exit 0;
diff --git a/egs/voxceleb/v2/local/prepare_for_eer.py b/egs/voxceleb/v2/local/prepare_for_eer.py
new file mode 100755
index 0000000000000000000000000000000000000000..2f569b70bc589c785943f95d28f3b2d72c8b20cc
--- /dev/null
+++ b/egs/voxceleb/v2/local/prepare_for_eer.py
@@ -0,0 +1,19 @@
+#!/usr/bin/env python3
+#
+# Copyright 2015 David Snyder
+# Apache 2.0.
+#
+# Copied from egs/sre10/v1/local/prepare_for_eer.py (commit 9cb4c4c2fb0223ee90c38d98af11305074eb7ef8)
+#
+# Given a trials and scores file, this script
+# prepares input for the binary compute-eer.
+import sys
+trials = open(sys.argv[1], 'r').readlines()
+scores = open(sys.argv[2], 'r').readlines()
+spkrutt2target = {}
+for line in trials:
+ spkr, utt, target = line.strip().split()
+ spkrutt2target[spkr+utt]=target
+for line in scores:
+ spkr, utt, score = line.strip().split()
+ print("{} {}".format(score, spkrutt2target[spkr+utt]))
diff --git a/egs/voxceleb/v2/local b/egs/voxceleb/v2/local2
similarity index 100%
rename from egs/voxceleb/v2/local
rename to egs/voxceleb/v2/local2
diff --git a/egs/voxforge/.DS_Store b/egs/voxforge/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..3ef0e7fb683034f86b60a4c47b17cdeeb7baaa4f
Binary files /dev/null and b/egs/voxforge/.DS_Store differ
diff --git a/egs/voxforge/s5/.DS_Store b/egs/voxforge/s5/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..fd02e9645239fa017d618ee71de4b79048707e6a
Binary files /dev/null and b/egs/voxforge/s5/.DS_Store differ
diff --git a/egs/wsj/.DS_Store b/egs/wsj/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..6ee6a89767d74d4211ea002f71327b2b72fb5169
Binary files /dev/null and b/egs/wsj/.DS_Store differ
diff --git a/src/.DS_Store b/src/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..a8186c21bbcc852b093943a73dda04ae082949ca
Binary files /dev/null and b/src/.DS_Store differ
diff --git a/src/matrix/t.bin b/src/matrix/t.bin
new file mode 100644
index 0000000000000000000000000000000000000000..36e7472b1e5339f1f2806d23959b4d102d9ac6c3
Binary files /dev/null and b/src/matrix/t.bin differ
diff --git a/tools/.DS_Store b/tools/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..d0424173c01af5637bc0771814c2a085e13868e8
Binary files /dev/null and b/tools/.DS_Store differ