Improve delimiter

engine/core/duplex_pipeline.py

@@ -632,6 +632,8 @@ class DuplexPipeline:
 
         split_idx = -1
         for idx, char in enumerate(text_buffer):
+            if char == "." and self._is_non_sentence_period(text_buffer, idx):
+                continue
             if char in self._SENTENCE_END_CHARS:
                 split_idx = idx
                 break
@@ -658,6 +660,29 @@ class DuplexPipeline:
         """Check whether text contains pronounceable content (not punctuation-only)."""
         return any(char.isalnum() for char in text)
 
+    def _is_non_sentence_period(self, text: str, idx: int) -> bool:
+        """Check whether '.' should NOT be treated as a sentence delimiter."""
+        if text[idx] != ".":
+            return False
+
+        # Decimal/version segment: 1.2, v1.2.3
+        if idx > 0 and idx < len(text) - 1 and text[idx - 1].isdigit() and text[idx + 1].isdigit():
+            return True
+
+        # Number abbreviations: No.1 / No. 1
+        left_start = idx - 1
+        while left_start >= 0 and text[left_start].isalpha():
+            left_start -= 1
+        left_token = text[left_start + 1:idx].lower()
+        if left_token == "no":
+            j = idx + 1
+            while j < len(text) and text[j].isspace():
+                j += 1
+            if j < len(text) and text[j].isdigit():
+                return True
+
+        return False
+
     async def _speak_sentence(self, text: str, fade_in_ms: int = 2, fade_out_ms: int = 8) -> None:
         """
         Synthesize and send a single sentence.

engine/services/siliconflow_tts.py

@@ -213,6 +213,29 @@ class StreamingTTSAdapter:
         self._cancel_event = asyncio.Event()
         self._is_speaking = False
 
+    def _is_non_sentence_period(self, text: str, idx: int) -> bool:
+        """Check whether '.' should NOT be treated as a sentence delimiter."""
+        if text[idx] != ".":
+            return False
+
+        # Decimal/version segment: 1.2, v1.2.3
+        if idx > 0 and idx < len(text) - 1 and text[idx - 1].isdigit() and text[idx + 1].isdigit():
+            return True
+
+        # Number abbreviations: No.1 / No. 1
+        left_start = idx - 1
+        while left_start >= 0 and text[left_start].isalpha():
+            left_start -= 1
+        left_token = text[left_start + 1:idx].lower()
+        if left_token == "no":
+            j = idx + 1
+            while j < len(text) and text[j].isspace():
+                j += 1
+            if j < len(text) and text[j].isdigit():
+                return True
+
+        return False
+    
     async def process_text_chunk(self, text_chunk: str) -> None:
         """
         Process a text chunk from LLM and trigger TTS when sentence is complete.
@@ -229,6 +252,8 @@ class StreamingTTSAdapter:
         while True:
             split_idx = -1
             for i, char in enumerate(self._buffer):
+                if char == "." and self._is_non_sentence_period(self._buffer, i):
+                    continue
                 if char in self.SENTENCE_ENDS:
                     split_idx = i
                     break