Fix sentence splitter to split on every sentence boundary

Previous version accumulated sentences until 500 chars, defeating
the purpose of streaming. Now splits on every sentence, only
combining very short sentences (<50 chars) with the next.

speech.py

@@ -517,38 +517,61 @@ def preprocess(raw_input):
     return raw_input
 
 def simple_sentence_split(text: str, max_length: int = 500) -> list[str]:
-    """Split text into sentences for better TTS processing.
+    """Split text into sentences for streaming TTS.
 
-    Simple sentence splitter that breaks on sentence boundaries (.!?)
+    Splits on sentence boundaries (.!?) for immediate streaming.
-    and ensures no sentence exceeds max_length characters.
+    Only combines very short sentences (< 50 chars) with the next one.
+    Long sentences exceeding max_length are split at word boundaries.
     """
-    # Split on sentence boundaries
+    # Split on sentence boundaries, keeping the punctuation
-    sentences = re.split(r'([.!?]+[\s\n]+)', text)
+    parts = re.split(r'([.!?]+\s*)', text)
 
-    # Recombine sentences with their punctuation
     result = []
     current = ""
 
-    for i in range(0, len(sentences), 2):
+    for i in range(0, len(parts), 2):
-        sentence = sentences[i]
+        sentence = parts[i].strip()
-        punct = sentences[i+1] if i+1 < len(sentences) else ""
+        punct = parts[i+1] if i+1 < len(parts) else ""
 
-        # If adding this sentence would exceed max_length, save current and start new
+        if not sentence:
-        if current and len(current) + len(sentence) + len(punct) > max_length:
+            continue
+
+        full_sentence = sentence + punct
+
+        # If sentence is very short, combine with current buffer
+        if len(full_sentence) < 50 and current:
+            current += " " + full_sentence
+        elif current:
+            # Save buffered content and start fresh
             result.append(current.strip())
-            current = sentence + punct
+            current = full_sentence
         else:
-            current += sentence + punct
+            current = full_sentence
 
     # Add remaining text
     if current.strip():
         result.append(current.strip())
 
-    # If we got nothing (no sentence boundaries), split on max_length
+    # Split any sentences that exceed max_length at word boundaries
-    if not result and text:
+    final_result = []
-        result = [text[i:i+max_length] for i in range(0, len(text), max_length)]
+    for sentence in result:
+        if len(sentence) <= max_length:
+            final_result.append(sentence)
+        else:
+            # Split at word boundaries
+            words = sentence.split()
+            chunk = ""
+            for word in words:
+                if len(chunk) + len(word) + 1 > max_length:
+                    if chunk:
+                        final_result.append(chunk.strip())
+                    chunk = word
+                else:
+                    chunk += " " + word if chunk else word
+            if chunk:
+                final_result.append(chunk.strip())
 
-    return result if result else [text]
+    return final_result if final_result else [text]
 
 # Auto-detect which model a voice belongs to (uses cached mapping)
 def detect_model_from_voice(voice: str) -> str: