fix scripts

examples/aishell-3/configs/v1.json

@@ -18,7 +18,7 @@
   },
   "data": {
     "max_wav_value": 32768.0,
-    "sampling_rate": 22050,
+    "sampling_rate": 44100,
     "filter_length": 1024,
     "hop_length": 256,
     "win_length": 1024,

examples/aishell-3/local/download_data.sh

@@ -29,6 +29,6 @@ fi
 
 cd $dir
 if ! tar -xvzf data_aishell3.tgz; then
-  echo "$0: error un-tarring archive $dir/$data_aishell3.tgz"
+  echo "$0: error un-tarring archive $dir/data_aishell3.tgz"
   exit 1;
 fi

examples/aishell-3/local/prepare_data.py

@@ -18,33 +18,34 @@
 import sys
 
 if len(sys.argv) != 4:
-    print('Usage: prepare_data.py lexicon in_data_dir out_data')
+    print("Usage: prepare_data.py lexicon in_data_dir out_data")
     sys.exit(-1)
 
 lexicon = {}
-with open(sys.argv[1], 'r', encoding='utf8') as fin:
+with open(sys.argv[1], "r", encoding="utf8") as fin:
     for line in fin:
         arr = line.strip().split()
         lexicon[arr[0]] = arr[1:]
 
-train_set_label_file = os.path.join(sys.argv[2], 'train',
-                                    'label_train-set.txt')
-with open(train_set_label_file, encoding='utf8') as fin, \
-     open(sys.argv[3], 'w', encoding='utf8') as fout:
+train_set_label_file = os.path.join(sys.argv[2], "train", "label_train-set.txt")
+with open(train_set_label_file, encoding="utf8") as fin, open(
+    sys.argv[3], "w", encoding="utf8"
+) as fout:
     # skip the first five lines in label_train-set.txt
     lines = [x.strip() for x in fin.readlines()][5:]
     for line in lines:
-        key, text, _ = line.split('|')
+        key, text, _ = line.split("|")
         speaker = key[:-4]
-        wav_path = os.path.join(sys.argv[2], 'train', 'wav', speaker,
-                                '{}.wav'.format(key))
+        wav_path = os.path.join(
+            sys.argv[2], "train", "wav", speaker, "{}.wav".format(key)
+        )
         phones = []
         for x in text.split():
-            if x == '%' or x == '$':
+            if x == "%" or x == "$":
                 phones.append(x)
             elif x in lexicon:
                 phones.extend(lexicon[x])
             else:
-                print('{} OOV {}'.format(key, x))
+                print("{} OOV {}".format(key, x))
                 sys.exit(-1)
-        fout.write('{}|{}|{}\n'.format(wav_path, speaker, ' '.join(phones)))
+        fout.write("{}|{}|sil {} sil\n".format(wav_path, speaker, " ".join(phones)))

examples/aishell-3/run.sh

@@ -8,17 +8,16 @@
 export CUDA_VISIBLE_DEVICES="0,1,2,3"
 
 stage=0  # start from -1 if you need to download data
-stop_stage=1
+stop_stage=3
 
 dataset_url=https://openslr.magicdatatech.com/resources/93/data_aishell3.tgz
 dataset_dir=. # path to dataset directory
 
-config=configs/v1.json
-dir=exp/v1  # training dir
-test_audio=test_audio
+dir=exp/v3  # training dir
+config=configs/v3.json
 
 data=data
-use_onnx=false
+test_audio=test_audio
 
 . tools/parse_options.sh || exit 1;
 
@@ -45,56 +44,52 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     sort | uniq | awk '{print $0, NR}' > $data/speaker.txt
   echo 'sil 0' > $data/phones.txt
   cat $data/all.txt | awk -F '|' '{print $3}' | \
-    awk '{ for (i=1;i<=NF;i++) print $i}' | \
-    sort | uniq | awk '{print $0, NR}' >> $data/phones.txt
+    awk '{for (i=1;i<=NF;i++) print $i}' | sort | uniq | \
+    grep -v 'sil' | awk '{print $0, NR}' >> $data/phones.txt
 
   # Split train/validation
-  cat $data/all.txt | shuf --random-source=<(yes 777) | head -n 110 | \
-    awk -F '|' '{print $1}' > $data/val.key
-  cat $data/all.txt | grep -f $data/val.key > $data/val.txt
-  head -10 $data/val.txt > $data/test.txt
-  sed -i '1,10d' $data/val.txt
-  cat $data/all.txt | grep -v -f $data/val.key > $data/train.txt
+  shuf --random-source=<(yes 777) $data/all.txt > $data/train.txt
+  head -n 100 $data/train.txt > $data/val.txt
+  sed -i '1,100d' $data/train.txt
+  head -n 10 $data/train.txt > $data/test.txt
+  sed -i '1,10d' $data/train.txt
 fi
 
 
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
   export MASTER_ADDR=localhost
-  export MASTER_PORT=10087
+  export MASTER_PORT=10086
   python vits/train.py -c $config -m $dir \
     --train_data $data/train.txt \
     --val_data $data/val.txt \
-    --phone_table $data/phones.txt \
-    --speaker_table $data/speaker.txt
+    --speaker_table $data/speaker.txt \
+    --phone_table $data/phones.txt
 fi
 
 
 if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
   mkdir -p $test_audio
-  python vits/inference.py  \
-    --checkpoint $dir/G_950000.pth --cfg $config \
-    --outdir $test_audio \
-    --phone_table $data/phones.txt \
+  python vits/inference.py --cfg $config \
     --speaker_table $data/speaker.txt \
-    --test_file $data/test.txt
+    --phone_table $data/phones.txt \
+    --checkpoint $dir/G_90000.pth \
+    --test_file $data/test.txt \
+    --outdir $test_audio
 fi
 
 
 if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
   mkdir -p $test_audio
-  python vits/export_onnx.py  \
-    --checkpoint $dir/G_950000.pth \
-    --cfg $config \
-    --onnx_model $dir/G_950000.onnx \
+  python vits/export_onnx.py --cfg $config \
+    --speaker_table $data/speaker.txt \
     --phone_table $data/phones.txt \
-    --speaker_table $data/speaker.txt
+    --checkpoint $dir/G_90000.pth \
+    --onnx_model $dir/G_90000.onnx
 
-  python vits/inference_onnx.py  \
-    --onnx_model $dir/G_950000.onnx \
-    --cfg $config \
-    --outdir $test_audio \
-    --phone_table $data/phones.txt \
+  python vits/inference_onnx.py --cfg $config \
     --speaker_table $data/speaker.txt \
-    --test_file $data/test.txt
+    --phone_table $data/phones.txt \
+    --onnx_model $dir/G_90000.onnx \
+    --test_file $data/test.txt \
+    --outdir $test_audio
 fi
-

examples/baker/local/prepare_data.py

@@ -3,40 +3,41 @@
 
 lexicon = {}
 
-with open(sys.argv[1], 'r', encoding='utf8') as fin:
+with open(sys.argv[1], "r", encoding="utf8") as fin:
     for line in fin:
         arr = line.strip().split()
         lexicon[arr[0]] = arr[1:]
 
-with open(sys.argv[2], 'r', encoding='utf8') as fin:
+with open(sys.argv[2], "r", encoding="utf8") as fin:
     lines = fin.readlines()
     for i in range(0, len(lines), 2):
         key = lines[i][:6]
         content = lines[i][7:].strip()
-        content = re.sub('[。，、“”？：……！（ ）—；]', '', content)
-        if 'Ｐ' in content:  # ignore utt 002365
+        content = re.sub("[。，、“”？：……！（ ）—；]", "", content)
+        if "Ｐ" in content:  # ignore utt 002365
             continue
         chars = []
         prosody = {}
 
         j = 0
         while j < len(content):
-            if content[j] == '#':
-                prosody[len(chars) - 1] = content[j:j + 2]
+            if content[j] == "#":
+                prosody[len(chars) - 1] = content[j : j + 2]
                 j += 2
             else:
                 chars.append(content[j])
                 j += 1
-        if key == '005107':
-            lines[i + 1] = lines[i + 1].replace(' ng1', ' en1')
+        if key == "005107":
+            lines[i + 1] = lines[i + 1].replace(" ng1", " en1")
         syllable = lines[i + 1].strip().split()
         s_index = 0
         phones = []
         for k, char in enumerate(chars):
             # 儿化音处理
             er_flag = False
-            if char == '儿' and (s_index == len(syllable)
-                                or syllable[s_index][0:2] != 'er'):
+            if char == "儿" and (
+                s_index == len(syllable) or syllable[s_index][0:2] != "er"
+            ):
                 er_flag = True
             else:
                 phones.extend(lexicon[syllable[s_index]])
@@ -47,5 +48,5 @@
                 else:
                     phones.append(prosody[k])
             else:
-                phones.append('#0')
-        print('{}/{}.wav|sil {} sil'.format(sys.argv[3], key, ' '.join(phones)))
+                phones.append("#0")
+        print("{}/{}.wav|baker|sil {} sil".format(sys.argv[3], key, " ".join(phones)))

examples/baker/run.sh

@@ -9,15 +9,14 @@ export CUDA_VISIBLE_DEVICES="0,1,2,3"  # specify your gpu id for training
 stage=0  # start from -1 if you need to download data
 stop_stage=3
 
-config=configs/v1.json  #
-dir=exp/v1  # training dir
-test_audio=test_audio
+dir=exp/v3  # training dir
+config=configs/v3.json
 
-# Please download data from https://www.data-baker.com/data/index/TNtts/, and
+# Please download data from https://www.data-baker.com/data/index/TNtts, and
 # set `raw_data_dir` to your data.
 raw_data_dir=/mnt/mnt-data-1/binbin.zhang/data/BZNSYP
 data=data
-use_onnx=false
+test_audio=test_audio
 
 . tools/parse_options.sh || exit 1;
 
@@ -33,19 +32,19 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     $raw_data_dir/ProsodyLabeling/000001-010000.txt \
     $raw_data_dir/Wave > $data/all.txt
 
-  echo 'baker 0' > $data/speaker.txt
-  echo 'sil 0' > $data/phones.txt
   cat $data/all.txt | awk -F '|' '{print $2}' | \
-    awk '{ for (i=1;i<=NF;i++) print $i}' | \
-    sort | uniq | awk '{print $0, NR}' >> $data/phones.txt
+    sort | uniq | awk '{print $0, NR}' > $data/speaker.txt
+  echo 'sil 0' > $data/phones.txt
+  cat $data/all.txt | awk -F '|' '{print $3}' | \
+    awk '{for (i=1;i<=NF;i++) print $i}' | sort | uniq | \
+    grep -v 'sil' | awk '{print $0, NR}' >> $data/phones.txt
 
   # Split train/validation
-  cat $data/all.txt | shuf --random-source=<(yes 777) | head -n 110 | \
-    awk -F '|' '{print $1}' > $data/val.key
-  cat $data/all.txt | grep -f $data/val.key > $data/val.txt
-  head -10 $data/val.txt > $data/test.txt
-  sed -i '1,10d' $data/val.txt
-  cat $data/all.txt | grep -v -f $data/val.key > $data/train.txt
+  shuf --random-source=<(yes 777) $data/all.txt > $data/train.txt
+  head -n 100 $data/train.txt > $data/val.txt
+  sed -i '1,100d' $data/train.txt
+  head -n 10 $data/train.txt > $data/test.txt
+  sed -i '1,10d' $data/train.txt
 fi
 
 
@@ -55,34 +54,34 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
   python vits/train.py -c $config -m $dir \
     --train_data $data/train.txt \
     --val_data $data/val.txt \
-    --phone_table $data/phones.txt \
-    --speaker_table $data/speaker.txt
+    --speaker_table $data/speaker.txt \
+    --phone_table $data/phones.txt
 fi
 
+
 if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
-  python vits/export_onnx.py  \
+  mkdir -p $test_audio
+  python vits/inference.py --cfg $config \
+    --speaker_table $data/speaker.txt \
+    --phone_table $data/phones.txt \
     --checkpoint $dir/G_90000.pth \
-    --cfg configs/base.json \
-    --onnx_model $dir/G_90000.onnx \
-    --phone_table data/phones.txt \
-    --speaker_table $data/speaker.txt
+    --test_file $data/test.txt \
+    --outdir $test_audio
 fi
 
+
 if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
-  [ ! -d ${test_audio} ] && mkdir ${test_audio}
-  if $use_onnx; then
-    python vits/inference_onnx.py  \
-      --onnx_model $dir/G_90000.onnx --cfg $config \
-      --outdir $test_audio \
-      --phone_table $data/phones.txt \
-      --test_file $data/test.txt \
-      --speaker_table $data/speaker.txt
-  else
-    python vits/inference.py  \
-      --checkpoint $dir/G_90000.pth --cfg $config \
-      --outdir $test_audio \
-      --phone_table $data/phones.txt \
-      --test_file $data/test.txt \
-      --speaker_table $data/speaker.txt
-  fi
+  mkdir -p $test_audio
+  python vits/export_onnx.py --cfg $config \
+    --speaker_table $data/speaker.txt \
+    --phone_table $data/phones.txt \
+    --checkpoint $dir/G_90000.pth \
+    --onnx_model $dir/G_90000.onnx
+
+  python vits/inference_onnx.py --cfg $config \
+    --speaker_table $data/speaker.txt \
+    --phone_table $data/phones.txt \
+    --onnx_model $dir/G_90000.onnx \
+    --test_file $data/test.txt \
+    --outdir $test_audio
 fi

examples/chinese_prosody_polyphone/frontend

@@ -0,0 +1 @@
+../../wetts/frontend