語料改口語調

fatchord-WaveRNN/dobi.yaml

@@ -8,7 +8,7 @@ meta:
 
 env=siatting:
   variables:
-  - PANPUN=0.2.1
+  - PANPUN=1.0
   - BANGTSI=https://www.dropbox.com/s/rye2sd0wo718bj5/SuiSiann-0.2.1.tar?dl=0
   - PINLUT=16k
 
@@ -182,17 +182,18 @@ job=wavernn-job:
 job=huatsiann-job:
   use: hunlian-khuanking
   env:
-  - KU={env.KU:tak10-ke7 tsə2-hue1 lai7 tsʰit8-tʰə5 !}
+  - KU={env.KU:to規-guân規 bûn規-huà本 beh再 siū規 lâng規 khíng規-tīng本.}
+#  - KU={env.KU:ta̍k規-ke規 tsò規-hué規 lâi規 tshit規-thô本!}
   - PANPUN={env.PANPUN}
-  - CUDA_VISIBLE_DEVICES=1
+  - CUDA_VISIBLE_DEVICES={env.CUDA_VISIBLE_DEVICES:0}
   command: |
     bash -eux -c '
       cp /tshamsoo/hparams.py .
       cp /tshamsoo/text_init.py utils/text/__init__.py
       cp /tshamsoo/text_symbols.py utils/text/symbols.py
       sed -i "s/required=True, //g" gen_tacotron.py
-      time python gen_tacotron.py --input_text "$KU" griffinlim
-      time python gen_tacotron.py --input_text "$KU" wavernn
-      time python gen_tacotron.py --input_text "$KU" wavernn --unbatched
+      time python gen_tacotron.py --input_text "$KU" --save_attention griffinlim
+      time python gen_tacotron.py --input_text "$KU" --save_attention wavernn
+      time python gen_tacotron.py --input_text "$KU" --save_attention wavernn --unbatched
     '
   mounts: [tshamsoo, checkpoints, outputs, ]

fatchord-WaveRNN/tshamsoo/preprocess-tacotron.py

@@ -14,12 +14,6 @@
 from utils.paths import Paths
 
 
-from 臺灣言語工具.解析整理.拆文分析器 import 拆文分析器
-from 臺灣言語工具.音標系統.閩南語.臺灣閩南語羅馬字拼音 import 臺灣閩南語羅馬字拼音
-from 臺灣言語工具.語音合成.閩南語音韻規則 import 閩南語音韻規則
-from 臺灣言語工具.語音合成 import 台灣話口語講法
-
-
 # Helper functions for argument types
 def valid_n_workers(num):
     n = int(num)
@@ -73,11 +67,7 @@ def suisiann(path: Union[str, Path], wav_files):
             mia = basename(tsua['音檔'])
             if mia in u_tihleh:
                 imtong = splitext(mia)[0]
-                hj = tsua['漢字']
-                lmj = tsua['羅馬字']
-                text_dict[imtong] = 台灣話口語講法(
-                    拆文分析器.建立句物件(hj, lmj)
-                )
+                text_dict[imtong] = tsua['口語調']
 
     return text_dict
 

fatchord-WaveRNN/tshamsoo/text_symbols.py

@@ -2,18 +2,24 @@
 
 '''
 Defines the set of symbols used in text input to the model.
-
-The default is a set of ASCII characters that works well for English or text that has been run through Unidecode. For other data, you can modify _characters. See TRAINING_DATA.md for details. '''
-from utils.text import cmudict
+The default is a set of ASCII characters that works well for English or text
+that has been run through Unidecode.
+See TRAINING_DATA.md for details.
+'''
 
 _pad = '_'
-_punctuation = '!\'(),.:;? '
-_special = '-'
-_letters = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz'
-_ipa = 'ŋəɛɨʔʰ̩ⁿ'
+_punctuation = r' -\.,;:?!"\'\(\)“”‘’~─'
+_letters = sorted(set(
+    'abcdefghijklmnopqrstuvwxyzáàâǎāa̍a̋éèêěēe̍e̋íìîǐīı̍i̍i̋'
+    'óòôǒōo̍őó͘ò͘ô͘ǒ͘ō͘o̍͘ő͘úùûǔūu̍űḿm̀m̂m̌m̄m̍m̋ńǹn̂ňn̄n̍n̋ⁿ'
+))
+_tsong = _letters + list(''.join(_letters).upper())
 
 _sooji = '0123456789'
+_piantiau = '規本隨再固三仔海'
 
 # Export all symbols:
-symbols = [_pad] + list(_special) + list(_punctuation) + list(_letters) + list(_ipa) + list(_sooji)
-
+symbols = (
+    [_pad] + list(_punctuation)
+    + list(_tsong) + list(_sooji) + list(_piantiau)
+)