diff --git a/ai-hub/app/api/routes/tts.py b/ai-hub/app/api/routes/tts.py
index 9c1d938..5352008 100644
--- a/ai-hub/app/api/routes/tts.py
+++ b/ai-hub/app/api/routes/tts.py
@@ -60,30 +60,55 @@
                 all_text = request.text
                 separators = ['.', '?', '!', '\n', '。', '？', '！', '，', ';']
                 
-                # Intelligent Multi-Splitter: Break text into a sequence of sentences/clauses
-                # instead of just "First" and "Rest".
-                def split_text(text, max_chunk_size=200):
+                # Adaptive Multi-Splitter: 
+                # 1. Tiny first chunk for instant response.
+                # 2. Larger later chunks to build a safety buffer for the stream.
+                def split_text(text):
                     chunks = []
-                    while len(text) > 0:
-                        if len(text) <= max_chunk_size:
-                            chunks.append(text)
+                    primary_seps = ['.', '?', '!', '\n', '。', '？', '！']
+                    secondary_seps = ['，', '；', ',', ';', ':']
+                    
+                    is_first = True
+                    current_text = text.strip()
+
+                    while current_text:
+                        # Target size: 80 for first (speed), 300 for rest (buffer)
+                        target_size = 80 if is_first else 300
+                        if len(current_text) <= target_size:
+                            chunks.append(current_text)
                             break
                         
-                        # Find the best separator to split at
                         split_at = -1
-                        # Look for a separator within a reasonable sentence length
-                        for i in range(min(len(text)-1, max_chunk_size), 20, -1):
-                            if text[i] in separators:
+                        # Priority 1: Sentence ends
+                        for i in range(min(len(current_text)-1, target_size), 40, -1):
+                            if current_text[i] in primary_seps:
                                 split_at = i + 1
                                 break
                         
+                        # Priority 2: Clauses/commas (only if no sentence end found)
+                        if split_at == -1:
+                            for i in range(min(len(current_text)-1, target_size), 40, -1):
+                                if current_text[i] in secondary_seps:
+                                    split_at = i + 1
+                                    break
+                                    
+                        # Priority 3: Spaces (forced)
+                        if split_at == -1:
+                            for i in range(min(len(current_text)-1, target_size), 40, -1):
+                                if current_text[i] == ' ':
+                                    split_at = i + 1
+                                    break
+
                         if split_at != -1:
-                            chunks.append(text[:split_at].strip())
-                            text = text[split_at:].strip()
+                            chunks.append(current_text[:split_at].strip())
+                            current_text = current_text[split_at:].strip()
+                            is_first = False
                         else:
-                            # Forced split if no separator found
-                            chunks.append(text[:max_chunk_size].strip())
-                            text = text[max_chunk_size:].strip()
+                            # Hard cutoff fallback
+                            chunks.append(current_text[:target_size].strip())
+                            current_text = current_text[target_size:].strip()
+                            is_first = False
+
                     return [c for c in chunks if c]
 
                 chunks = split_text(all_text)