diff --git a/backend/app/services/gemini_tts.py b/backend/app/services/gemini_tts.py
index c30b98e..1cf3c93 100644
--- a/backend/app/services/gemini_tts.py
+++ b/backend/app/services/gemini_tts.py
@@ -1,4 +1,5 @@
 import io
+import re
 
 from google import genai
 from google.genai import types
@@ -30,6 +31,18 @@ class GeminiTTSService:
         self.default_voice = settings.gemini_tts_default_voice
         logger.info(f"Gemini TTS service initialized with model: {self.model}")
 
+    @staticmethod
+    def _extract_retry_after(error: Exception) -> float | None:
+        """Return seconds to wait from a Google API 429 retryDelay, or None."""
+        msg = str(error)
+        m = re.search(r"retry in ([0-9.]+)s", msg, re.IGNORECASE)
+        if m:
+            return float(m.group(1)) + 5
+        m = re.search(r"'retryDelay':\s*'([0-9.]+)s'", msg)
+        if m:
+            return float(m.group(1)) + 5
+        return None
+
     async def synthesize_text(
         self,
         text: str,
@@ -234,8 +247,9 @@ class GeminiTTSService:
                 api_response_info = str(e)
 
                 if attempt < max_attempts - 1:
-                    # Exponential backoff with jitter
-                    delay = base_delay * (2 ** attempt) + random.uniform(0, 1)
+                    # Honour API-provided retry delay on 429; fall back to exponential backoff
+                    api_delay = self._extract_retry_after(e)
+                    delay = api_delay if api_delay else base_delay * (2 ** attempt) + random.uniform(0, 1)
                     logger.warning(
                         f"TTS synthesis attempt {attempt + 1}/{max_attempts} failed for cue {cue_index}. "
                         f"Retrying in {delay:.2f}s. Error: {e}"
diff --git a/backend/app/tasks/tts_synthesis.py b/backend/app/tasks/tts_synthesis.py
index 0f97320..658b4fc 100644
--- a/backend/app/tasks/tts_synthesis.py
+++ b/backend/app/tasks/tts_synthesis.py
@@ -8,6 +8,7 @@ in parallel using a dedicated TTS worker with concurrency=8.
 import asyncio
 import hashlib
 import io
+import re
 import time
 
 from celery import group
@@ -23,6 +24,21 @@ from . import celery_app
 
 logger = get_logger(__name__)
 
+
+def _extract_retry_after(error: Exception) -> float | None:
+    """Return seconds to wait from a Google API 429 retryDelay, or None."""
+    msg = str(error)
+    # "Please retry in 37.65s" pattern from the message text
+    m = re.search(r"retry in ([0-9.]+)s", msg, re.IGNORECASE)
+    if m:
+        return float(m.group(1)) + 5
+    # 'retryDelay': '37s' pattern in the JSON body
+    m = re.search(r"'retryDelay':\s*'([0-9.]+)s'", msg)
+    if m:
+        return float(m.group(1)) + 5
+    return None
+
+
 _TTS_PROVIDER_MODEL_MAP = {
     # (provider, model) → cost-tracker provider + model strings
     "gemini": "google",
@@ -169,9 +185,10 @@ def synthesize_cue_task(
 
         # Check if we have retries left
         if self.request.retries < self.max_retries:
-            # Calculate backoff delay with jitter
             import random
-            delay = (2 ** self.request.retries) + random.uniform(0, 1)
+            # Honour the API-provided retry delay on 429; fall back to exponential backoff
+            api_delay = _extract_retry_after(e)
+            delay = api_delay if api_delay else (2 ** self.request.retries) + random.uniform(0, 1)
             logger.info(
                 f"Retrying TTS cue {cue_index} in {delay:.1f}s "
                 f"(attempt {self.request.retries + 2}/{self.max_retries + 1})"