AMI hotfix

egs/ami/s5b/local/ami_download.sh

@@ -51,13 +51,17 @@ mkdir -p $wdir/log
 
 #download waves
 
-cat local/split_train.orig local/split_eval.orig local/split_dev.orig > $wdir/ami_meet_ids.flist
-
 wgetfile=$wdir/wget_$mic.sh
 
-# TODO fix this with Pawel, files don't exist anymore,
-manifest="wget --continue -O $adir/MANIFEST.TXT http://groups.inf.ed.ac.uk/ami/download/temp/amiBuild-04237-Sun-Jun-15-2014.manifest.txt"
-license="wget --continue -O $adir/LICENCE.TXT http://groups.inf.ed.ac.uk/ami/download/temp/Creative-Commons-Attribution-NonCommercial-ShareAlike-2.5.txt"
+cp local/MANIFEST.TXT $adir/MANIFEST.TXT
+manifest=$adir/MANIFEST.TXT
+#manifest="wget --continue -O $adir/MANIFEST.TXT https://groups.inf.ed.ac.uk/ami/download/temp/amiBuild-1372-Thu-Apr-28-2022.manifest.txt"
+license="wget --continue -O $adir/LICENSE.TXT http://groups.inf.ed.ac.uk/ami/corpus/license.shtml"
+
+# Parse the manifest file, and separate recordings into train, dev, and eval sets
+# python3 local/split_manifest.py $adir/MANIFEST.TXT
+
+cat local/split_train.orig local/split_eval.orig local/split_dev.orig > $wdir/ami_meet_ids.flist
 
 echo "#!/usr/bin/env bash" > $wgetfile
 echo $manifest >> $wgetfile
@@ -85,23 +89,6 @@ echo "Downloading audio files for $mic scenario."
 echo "Look at $wdir/log/download_ami_$mic.log for progress"
 $wgetfile &> $wdir/log/download_ami_$mic.log
 
-# Do rough check if #wavs is as expected, it will fail anyway in data prep stage if it isn't,
-if [ "$mic" == "ihm" ]; then
-  num_files=$(find $adir -iname *Headset* | wc -l)
-  if [ $num_files -ne 687 ]; then
-    echo "Warning: Found $num_files headset wavs but expected 687. Check $wdir/log/download_ami_$mic.log for details."
-    exit 1;
-  fi
-else
-  num_files=$(find $adir -iname *Array1* | wc -l)
-  if [[ $num_files -lt 1352 && "$mic" == "mdm" ]]; then
-    echo "Warning: Found $num_files distant Array1 waves but expected 1352 for mdm. Check $wdir/log/download_ami_$mic.log for details."
-    exit 1;
-  elif [[ $num_files -lt 169 && "$mic" == "sdm" ]]; then
-    echo "Warning: Found $num_files distant Array1 waves but expected 169 for sdm. Check $wdir/log/download_ami_$mic.log for details."
-    exit 1;
-  fi
-fi
 
 echo "Downloads of AMI corpus completed succesfully. License can be found under $adir/LICENCE.TXT"
 exit 0;

egs/ami/s5c/cmd.sh

@@ -10,6 +10,6 @@
 # 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 train_cmd="queue.pl --mem 4G"
-export decode_cmd="queue.pl --mem 4G"
+export train_cmd="run.pl --mem 4G"
+export decode_cmd="run.pl --mem 4G"
 

egs/ami/s5c/local/nnet3/xvector/tuning/run_xvector_1a.sh

@@ -54,19 +54,18 @@ num_pdfs=$(awk '{print $2}' $data/utt2spk | sort | uniq -c | wc -l)
 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/callhome_diarization/v2/xvector-$(date +'%m_%d_%H_%M')/$egs_dir/storage $egs_dir/storage
-  fi
+  # frame per iter original 1000000000
+  # frame per iter diagnostic original 500000
+  # num repeat original 1
   sid/nnet3/xvector/get_egs.sh --cmd "$train_cmd" \
     --nj 8 \
     --stage 0 \
-    --frames-per-iter 1000000000 \
-    --frames-per-iter-diagnostic 500000 \
+    --frames-per-iter 100000 \
+    --frames-per-iter-diagnostic 10000 \
     --min-frames-per-chunk 200 \
     --max-frames-per-chunk 400 \
     --num-diagnostic-archives 3 \
-    --num-repeats 40 \
+    --num-repeats 10 \
     "$data" $egs_dir
 fi
 

egs/ami/s5c/local/prepare_data.py

@@ -20,14 +20,16 @@
 
 def find_audios(wav_path, file_list):
     # Get all wav file names from audio directory
-    command = 'find %s -name "*Mix-Headset.wav"' % (wav_path)
+    command = 'find %s -name "*.wav"' % (wav_path)
     wavs = subprocess.check_output(command, shell=True).decode('utf-8').splitlines()
     keys = [ os.path.splitext(os.path.basename(wav))[0] for wav in wavs ]
     data = {'key': keys, 'file_path': wavs}
     df_wav = pd.DataFrame(data)
 
     # Filter list to keep only those in annotations (for the specific data split)
     file_names_str = "|".join(file_list)
+    print(file_names_str)
+    print(df_wav)
     df_wav = df_wav.loc[df_wav['key'].str.contains(file_names_str)].sort_values('key')
     return df_wav
 

egs/ami/s5c/local/split_dev.orig

@@ -0,0 +1,8 @@
+ES2011a
+ES2011b
+ES2011c
+ES2011d
+IB4001
+IB4002
+IB4003
+IB4004

egs/ami/s5c/local/split_eval.orig

@@ -0,0 +1,8 @@
+ES2004a
+ES2004b
+ES2004c
+ES2004d
+EN2002a
+EN2002b
+EN2002c
+EN2002d

egs/ami/s5c/local/split_manifest.py

@@ -0,0 +1,69 @@
+import os
+import sys
+
+def unique(m):
+    unique_list = []
+
+    for i in m:
+        if i not in unique_list:
+            unique_list.append(i)
+
+    return unique_list
+
+# Load in the MANIFEST file, save off the audio recoding file names
+file = sys.argv[1]
+prefix = '\thttps://groups.inf.ed.ac.uk/ami/AMICorpusMirror//amicorpus/'
+m = []
+
+with open(file) as f:
+    for line in f:
+        #splits = line.split('/')
+        #print(splits)
+        if line.startswith(prefix):
+            splits = line.split('/')
+            #print(splits)
+            m.append(splits[7])
+m = unique(m)
+print("Got the audio files from MANIFEST.TXT")
+#print(m)
+
+# Separate files and save off into train, dev, and eval partitions
+N = len(m)
+
+#train = m[:round(N*.5)]
+#dev = m[round(N*.5)+1:round(N*.8)]
+#ev = m[round(N*.8)+1:]
+
+#train = train[:12]
+#dev = dev[:10]
+#ev = ev[:10]
+
+train = m[:8]
+dev = m[9:15]
+ev = m[16:20]
+
+print("Train set: "+str(train))
+print("Dev set: "+str(dev))
+print("Eval set: "+str(ev))
+
+if os.path.exists('local/split_train.orig'):
+    os.remove('local/split_train.orig')
+if os.path.exists('local/split_dev.orig'):
+    os.remove('local/split_dev.orig')
+if os.path.exists('local/split_eval.orig'):
+    os.remove('local/split_eval.orig')
+
+with open('local/split_train.orig', 'a') as train_file:
+    for d in train:
+        train_file.write(d)
+        train_file.write("\n")
+
+with open('local/split_dev.orig', 'a') as dev_file:
+    for d in dev:
+        dev_file.write(d)
+        dev_file.write("\n")
+
+with open('local/split_eval.orig', 'a') as eval_file:
+    for d in ev:
+        eval_file.write(d)
+        eval_file.write("\n")

egs/ami/s5c/local/split_train.orig

@@ -0,0 +1,27 @@
+IS1000a
+IS1000b
+IS1000c
+IS1000d
+IS1001a
+IS1001b
+IS1001c
+IS1001d
+IS1002b
+IS1002c
+IS1002d
+ES2002a
+ES2002b
+ES2002c
+ES2002d
+ES2003a
+ES2003b
+ES2003c
+ES2003d
+ES2005a
+ES2005b
+ES2005c
+ES2005d
+ES2006a
+ES2006b
+ES2006c
+ES2006d

egs/ami/s5c/run.sh

@@ -20,12 +20,14 @@
 set -euo pipefail
 mfccdir=`pwd`/mfcc
 
-stage=0
+stage=7
 overlap_stage=0
 diarizer_stage=0
-nj=50
+nj=10
 decode_nj=15
 
+export mic=ihm
+
 model_dir=exp/xvector_nnet_1a
 
 train_set=train
@@ -37,11 +39,6 @@ diarizer_type=spectral  # must be one of (ahc, spectral, vbx)
 
 # Path where AMI gets downloaded (or where locally available):
 AMI_DIR=$PWD/wav_db # Default,
-case $(hostname -d) in
-  fit.vutbr.cz) AMI_DIR=/mnt/matylda5/iveselyk/KALDI_AMI_WAV ;; # BUT,
-  clsp.jhu.edu) AMI_DIR=/export/corpora5/amicorpus ;; # JHU,
-  cstr.ed.ac.uk) AMI_DIR= ;; # Edinburgh,
-esac
 
 # Download AMI corpus, You need around 130GB of free space to get whole data
 if [ $stage -le 1 ]; then
@@ -87,13 +84,16 @@ if [ $stage -le 3 ]; then
     steps/make_mfcc.sh --mfcc-config conf/mfcc_hires.conf --nj $nj --cmd "$train_cmd" data/$dataset
     steps/compute_cmvn_stats.sh data/$dataset
     utils/fix_data_dir.sh data/$dataset
+    echo "FEATURES COMPLETE FOR DATASET"
   done
 fi
 
 if [ $stage -le 4 ]; then
   echo "$0: preparing a AMI training data to train PLDA model"
-  local/nnet3/xvector/prepare_feats.sh --nj $nj --cmd "$train_cmd" \
-    data/train data/plda_train exp/plda_train_cmn
+  #local/nnet3/xvector/prepare_feats.sh --nj $nj --cmd "$train_cmd" \
+  #  data/train data/plda_train exp/plda_train_cmn
+  local/nnet3/xvector/run_xvector.sh --stage $stage --train-stage -1 \
+    --data data/plda_train
 fi
 
 if [ $stage -le 5 ]; then

egs/ami/s5c/sid

@@ -0,0 +1 @@
+../../sre08/v1/sid

egs/ami/s5c_apt2141/README.txt

@@ -0,0 +1,38 @@
+A. Alexander Thornton (apt2141)
+
+B. May 8, 2022
+
+C. Project Title: Speaker Diarization: Deep Speech Embeddings for Time Delay Neural Networks (TDNN)
+
+D. Project Summary:
+
+Abstract—The fundamental problem of Speaker Diarization
+can be simplified as ”who spoke when”. At its essence, Speaker
+Diarization can be reduced to the traditional Speaker Identifica-
+tion problem, but expanded to N interleaving speakers through
+time. This work improves upon the existing Speaker Diarization
+project in Kaldi, which was incomplete and unfinished prior to
+my efforts.
+Index Terms—speaker identification, diarization, time delay
+neural networks, time series learning
+
+E. All tools are included with the code here. Build Kaldi, and you can just run the run.sh
+
+F. Only use run.sh, the stage is set to 7 for decoding
+
+G. Run the code with this simple command:
+
+	./run.sh
+
+All environment variables are defined inside
+
+Sample output will appear at the bottom, with the test accuracy
+
+H. The data used was built off a MANIFEXT file downloaded here:
+
+https://groups.inf.ed.ac.uk/ami/download/temp/amiBuild-1372-Thu-Apr-28-2022.manifest.txt
+
+It's important to know that those files change daily, and are constantly changing, so this one might already be gone
+
+
+

egs/ami/s5c_apt2141/cmd.sh

@@ -0,0 +1,15 @@
+# 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 train_cmd="run.pl --mem 4G"
+export decode_cmd="run.pl --mem 4G"
+

egs/ami/s5c_apt2141/conf/decode.conf

@@ -0,0 +1,3 @@
+beam=11.0 # beam for decoding.  Was 13.0 in the scripts.
+first_beam=8.0 # beam for 1st-pass decoding in SAT.
+

egs/ami/s5c_apt2141/conf/mfcc.conf

@@ -0,0 +1,2 @@
+--use-energy=false   # only non-default option.
+--sample-frequency=16000

egs/ami/s5c_apt2141/conf/mfcc_hires.conf

@@ -0,0 +1,10 @@
+# config for high-resolution MFCC features, intended for neural network training
+# Note: we keep all cepstra, so it has the same info as filterbank features,
+# but MFCC is more easily compressible (because less correlated) which is why 
+# we prefer this method.
+--use-energy=false   # use average of log energy, not energy.
+--num-mel-bins=40     # similar to Google's setup.
+--num-ceps=40     # there is no dimensionality reduction.
+--low-freq=20     # low cutoff frequency for mel bins... this is high-bandwidth data, so
+                  # there might be some information at the low end.
+--high-freq=-400 # high cutoff frequently, relative to Nyquist of 8000 (=7600) 

egs/ami/s5c_apt2141/conf/online_cmvn.conf

@@ -0,0 +1,3 @@
+# configuration file for apply-cmvn-online, used in the script ../local/run_online_decoding.sh
+--norm-means=true
+--norm-vars=false

egs/ami/s5c_apt2141/conf/pitch.conf

@@ -0,0 +1 @@
+--sample-frequency=16000

egs/ami/s5c_apt2141/diarization

@@ -0,0 +1 @@
+../../callhome_diarization/v1/diarization