Merge branch 'RVC-Boss:main' into dev

GPT_SoVITS/text/cantonese.py

@@ -3,8 +3,8 @@
 import sys
 import re
 import cn2an
+import ToJyutping
 
-from pyjyutping import jyutping
 from text.symbols import punctuation
 from text.zh_normalization.text_normlization import TextNormalizer
 
@@ -173,12 +173,24 @@ def jyuping_to_initials_finals_tones(jyuping_syllables):
 
 
 def get_jyutping(text):
-    jp = jyutping.convert(text)
-    # print(1111111,jp)
-    for symbol in punctuation:
-        jp = jp.replace(symbol, " " + symbol + " ")
-    jp_array = jp.split()
-    return jp_array
+    jyutping_array = []
+    punct_pattern = re.compile(r"^[{}]+$".format(re.escape("".join(punctuation))))
+
+    syllables = ToJyutping.get_jyutping_list(text)
+
+    for word, syllable in syllables:
+        if punct_pattern.match(word):
+            puncts = re.split(r"([{}])".format(re.escape("".join(punctuation))), word)
+            for punct in puncts:
+                if len(punct) > 0:
+                    jyutping_array.append(punct)
+        else:
+            # match multple jyutping eg: liu4 ge3, or single jyutping eg: liu4
+            if not re.search(r"^([a-z]+[1-6]+[ ]?)+$", syllable):
+                raise ValueError(f"Failed to convert {word} to jyutping: {syllable}")
+            jyutping_array.append(syllable)
+
+    return jyutping_array
 
 
 def get_bert_feature(text, word2ph):

GPT_SoVITS/text/g2pw/g2pw.py

@@ -144,11 +144,16 @@ def read_dict():
     return polyphonic_dict
 
 
-def correct_pronunciation(word,word_pinyins):
-    if word in pp_dict:
-        word_pinyins = pp_dict[word]
-
-    return word_pinyins
+def correct_pronunciation(word, word_pinyins):
+    new_pinyins = pp_dict.get(word, "")
+    if new_pinyins == "":
+        for idx, w in enumerate(word):
+            w_pinyin = pp_dict.get(w, "")
+            if w_pinyin != "":
+                word_pinyins[idx] = w_pinyin[0]
+        return word_pinyins
+    else:
+        return new_pinyins
 
 
 pp_dict = get_dict()

requirements.txt

@@ -27,7 +27,7 @@ LangSegment>=0.2.0
 Faster_Whisper
 wordsegment
 rotary_embedding_torch
-pyjyutping 
+ToJyutping 
 g2pk2
 ko_pron
 opencc; sys_platform != 'linux'

webui.py

@@ -72,7 +72,7 @@ mem = []
 if_gpu_ok = False
 
 # 判断是否有能用来训练和加速推理的N卡
-ok_gpu_keywords={"10","16","20","30","40","A2","A3","A4","P4","A50","500","A60","70","80","90","M4","T4","TITAN","L4","4060","H"}
+ok_gpu_keywords={"10","16","20","30","40","A2","A3","A4","P4","A50","500","A60","70","80","90","M4","T4","TITAN","L4","4060","H","600"}
 set_gpu_numbers=set()
 if torch.cuda.is_available() or ngpu != 0:
     for i in range(ngpu):