From aa6ac3e63b659c4574ec45985fb3757b59355567 Mon Sep 17 00:00:00 2001
From: liyulingyue <852433440@qq.com>
Date: Fri, 17 Jan 2025 07:43:29 +0800
Subject: [PATCH] fit with pypinyin

---
 paddlespeech/t2s/frontend/tone_sandhi.py | 31 ++++++++++++++++--------
 paddlespeech/t2s/frontend/zh_frontend.py |  5 ++++
 2 files changed, 26 insertions(+), 10 deletions(-)

diff --git a/paddlespeech/t2s/frontend/tone_sandhi.py b/paddlespeech/t2s/frontend/tone_sandhi.py
index 3558064cd..ec1b61302 100644
--- a/paddlespeech/t2s/frontend/tone_sandhi.py
+++ b/paddlespeech/t2s/frontend/tone_sandhi.py
@@ -243,8 +243,10 @@ class ToneSandhi():
             if skip_next:
                 skip_next = False
                 continue
-            if i - 1 >= 0 and word == "一" and i + 1 < len(seg) and seg[i - 1][0] == seg[i + 1][0] and seg[i - 1][1] == "v":
-                new_seg[-1] = (new_seg[-1][0] + "一" + seg[i + 1][0], new_seg[-1][1])
+            if i - 1 >= 0 and word == "一" and i + 1 < len(seg) and seg[i - 1][
+                    0] == seg[i + 1][0] and seg[i - 1][1] == "v":
+                new_seg[-1] = (new_seg[-1][0] + "一" + seg[i + 1][0],
+                               new_seg[-1][1])
                 skip_next = True
             else:
                 new_seg.append((word, pos))
@@ -262,11 +264,16 @@ class ToneSandhi():
     def _merge_continuous_three_tones(
             self, seg: List[Tuple[str, str]]) -> List[Tuple[str, str]]:
         new_seg = []
-        sub_finals_list = [
-            lazy_pinyin(
+        sub_finals_list = []
+        for (word, pos) in seg:
+            orig_finals = lazy_pinyin(
                 word, neutral_tone_with_five=True, style=Style.FINALS_TONE3)
-            for (word, pos) in seg
-        ]
+            # after pypinyin==0.41.0, 嗯 will be processed as "", so we need to change it to "n2"
+            en_index = [index for index, c in enumerate(word) if c == "嗯"]
+            for i in en_index:
+                orig_finals[i] = "n2"
+            sub_finals_list.append(orig_finals)
+
         assert len(sub_finals_list) == len(seg)
         merge_last = [False] * len(seg)
         for i, (word, pos) in enumerate(seg):
@@ -292,11 +299,15 @@ class ToneSandhi():
     def _merge_continuous_three_tones_2(
             self, seg: List[Tuple[str, str]]) -> List[Tuple[str, str]]:
         new_seg = []
-        sub_finals_list = [
-            lazy_pinyin(
+        sub_finals_list = []
+        for (word, pos) in seg:
+            orig_finals = lazy_pinyin(
                 word, neutral_tone_with_five=True, style=Style.FINALS_TONE3)
-            for (word, pos) in seg
-        ]
+            # after pypinyin==0.41.0, 嗯 will be processed as "", so we need to change it to "n2"
+            en_index = [index for index, c in enumerate(word) if c == "嗯"]
+            for i in en_index:
+                orig_finals[i] = "n2"
+            sub_finals_list.append(orig_finals)
         assert len(sub_finals_list) == len(seg)
         merge_last = [False] * len(seg)
         for i, (word, pos) in enumerate(seg):
diff --git a/paddlespeech/t2s/frontend/zh_frontend.py b/paddlespeech/t2s/frontend/zh_frontend.py
index 1431bc6d8..505fd4699 100644
--- a/paddlespeech/t2s/frontend/zh_frontend.py
+++ b/paddlespeech/t2s/frontend/zh_frontend.py
@@ -173,6 +173,11 @@ class Frontend():
                 word, neutral_tone_with_five=True, style=Style.INITIALS)
             orig_finals = lazy_pinyin(
                 word, neutral_tone_with_five=True, style=Style.FINALS_TONE3)
+            # after pypinyin==0.41.0, 嗯 will be processed as "", so we need to change it to "n2"
+            en_index = [index for index, c in enumerate(word) if c == "嗯"]
+            for i in en_index:
+                orig_finals[i] = "n2"
+
             for c, v in zip(orig_initials, orig_finals):
                 if re.match(r'i\d', v):
                     if c in ['z', 'c', 's']: