diff --git a/changelog/3946.added.md b/changelog/3946.added.md
new file mode 100644
index 000000000..6aabfefb2
--- /dev/null
+++ b/changelog/3946.added.md
@@ -0,0 +1 @@
+- Runtime settings updates (via `STTUpdateSettingsFrame`) now work for AWS Transcribe, Azure, Cartesia, Deepgram, ElevenLabs Realtime, Gradium, and Soniox STT services. Previously, changing settings at runtime only stored the new values without reconnecting.
diff --git a/src/pipecat/services/aws/stt.py b/src/pipecat/services/aws/stt.py
index ca3b0140c..7355bb1b7 100644
--- a/src/pipecat/services/aws/stt.py
+++ b/src/pipecat/services/aws/stt.py
@@ -107,7 +107,7 @@ class AWSTranscribeSTTService(WebsocketSTTService):
             _warn_deprecated_param("language", AWSTranscribeSTTSettings, "language")
             default_settings.language = self.language_to_service_language(language)
 
-        # 3. No params to apply
+        # 3. (No step 3, as there's no params object to apply)
 
         # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
@@ -161,7 +161,7 @@ class AWSTranscribeSTTService(WebsocketSTTService):
         """Apply a settings delta and reconnect if anything changed."""
         changed = await super()._update_settings(delta)
 
-        if changed:
+        if changed and self._websocket:
             await self._disconnect()
             await self._connect()
 
diff --git a/src/pipecat/services/azure/stt.py b/src/pipecat/services/azure/stt.py
index f66cfeb42..f940ea9c0 100644
--- a/src/pipecat/services/azure/stt.py
+++ b/src/pipecat/services/azure/stt.py
@@ -112,7 +112,7 @@ class AzureSTTService(STTService):
             _warn_deprecated_param("language", AzureSTTSettings, "language")
             default_settings.language = language_to_azure_language(language)
 
-        # 3. No params to apply
+        # 3. (No step 3, as there's no params object to apply)
 
         # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
@@ -166,8 +166,9 @@ class AzureSTTService(STTService):
             self._speech_config.speech_recognition_language = (
                 self._settings.language or language_to_azure_language(Language.EN_US)
             )
-            await self._disconnect()
-            await self._connect()
+            if self._audio_stream:
+                await self._disconnect()
+                await self._connect()
 
         return changed
 
diff --git a/src/pipecat/services/deepgram/stt.py b/src/pipecat/services/deepgram/stt.py
index 17570911a..14020631b 100644
--- a/src/pipecat/services/deepgram/stt.py
+++ b/src/pipecat/services/deepgram/stt.py
@@ -365,7 +365,9 @@ class DeepgramSTTService(STTService):
             vad_events=False,
         )
 
-        # 2. Apply live_options overrides — only if settings not provided
+        # 2. (No step 2, as there are no deprecated direct args)
+
+        # 3. Apply live_options overrides — only if settings not provided
         if live_options is not None:
             _warn_deprecated_param("live_options", DeepgramSTTSettings)
             if not settings:
@@ -402,7 +404,7 @@ class DeepgramSTTService(STTService):
                 delta = DeepgramSTTSettings.from_mapping(lo_dict)
                 default_settings.apply_update(delta)
 
-        # 3. Apply settings delta (canonical API, always wins)
+        # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
             default_settings.apply_update(settings)
 
@@ -494,8 +496,9 @@ class DeepgramSTTService(STTService):
         if isinstance(self._settings, DeepgramSTTSettings):
             self._settings._sync_extra_to_fields()
 
-        await self._disconnect()
-        await self._connect()
+        if self._connection:
+            await self._disconnect()
+            await self._connect()
 
         return changed
 
diff --git a/src/pipecat/services/deepgram/tts.py b/src/pipecat/services/deepgram/tts.py
index 7749fc370..5d6e5ffdc 100644
--- a/src/pipecat/services/deepgram/tts.py
+++ b/src/pipecat/services/deepgram/tts.py
@@ -110,6 +110,8 @@ class DeepgramTTSService(WebsocketTTSService):
             default_settings.model = voice
             default_settings.voice = voice
 
+        # 3. (No step 3, as there's no params object to apply)
+
         # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
             default_settings.apply_update(settings)
@@ -423,6 +425,8 @@ class DeepgramHttpTTSService(TTSService):
             default_settings.model = voice
             default_settings.voice = voice
 
+        # 3. (No step 3, as there's no params object to apply)
+
         # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
             default_settings.apply_update(settings)
diff --git a/src/pipecat/services/elevenlabs/stt.py b/src/pipecat/services/elevenlabs/stt.py
index 1802bb095..dff501d22 100644
--- a/src/pipecat/services/elevenlabs/stt.py
+++ b/src/pipecat/services/elevenlabs/stt.py
@@ -605,8 +605,9 @@ class ElevenLabsRealtimeSTTService(WebsocketSTTService):
         if not changed:
             return changed
 
-        await self._disconnect()
-        await self._connect()
+        if self._websocket:
+            await self._disconnect()
+            await self._connect()
 
         return changed
 
diff --git a/src/pipecat/services/gradium/stt.py b/src/pipecat/services/gradium/stt.py
index 05a6bb8b5..814d478e4 100644
--- a/src/pipecat/services/gradium/stt.py
+++ b/src/pipecat/services/gradium/stt.py
@@ -12,7 +12,7 @@ WebSocket API for streaming audio transcription.
 
 import base64
 import json
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Any, AsyncGenerator, Optional
 
 from loguru import logger
@@ -28,7 +28,7 @@ from pipecat.frames.frames import (
     VADUserStoppedSpeakingFrame,
 )
 from pipecat.processors.frame_processor import FrameDirection
-from pipecat.services.settings import STTSettings, _warn_deprecated_param
+from pipecat.services.settings import NOT_GIVEN, STTSettings, _NotGiven, _warn_deprecated_param
 from pipecat.services.stt_latency import GRADIUM_TTFS_P99
 from pipecat.services.stt_service import WebsocketSTTService
 from pipecat.transcriptions.language import Language, resolve_language
@@ -77,7 +77,7 @@ class GradiumSTTSettings(STTSettings):
             Default is 10 (800ms). Lower values like 7-8 give faster response.
     """
 
-    delay_in_frames: Optional[int] = None
+    delay_in_frames: Optional[int] | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
 
 
 class GradiumSTTService(WebsocketSTTService):
@@ -157,7 +157,9 @@ class GradiumSTTService(WebsocketSTTService):
             delay_in_frames=None,
         )
 
-        # 2. Apply params overrides — only if settings not provided
+        # 2. (No step 2, as there are no deprecated direct args)
+
+        # 3. Apply params overrides — only if settings not provided
         if params is not None:
             _warn_deprecated_param("params", GradiumSTTSettings)
             if not settings:
@@ -165,7 +167,7 @@ class GradiumSTTService(WebsocketSTTService):
                 if params.delay_in_frames is not None:
                     default_settings.delay_in_frames = params.delay_in_frames
 
-        # 3. Apply settings delta (canonical API, always wins)
+        # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
             default_settings.apply_update(settings)
 
@@ -213,8 +215,9 @@ class GradiumSTTService(WebsocketSTTService):
         if not changed:
             return changed
 
-        await self._disconnect()
-        await self._connect()
+        if self._websocket:
+            await self._disconnect()
+            await self._connect()
         return changed
 
     async def start(self, frame: StartFrame):
diff --git a/src/pipecat/services/lmnt/tts.py b/src/pipecat/services/lmnt/tts.py
index 3cc6778af..f6bf46649 100644
--- a/src/pipecat/services/lmnt/tts.py
+++ b/src/pipecat/services/lmnt/tts.py
@@ -139,7 +139,7 @@ class LmntTTSService(InterruptibleTTSService):
             _warn_deprecated_param("model", LmntTTSSettings, "model")
             default_settings.model = model
 
-        # 3. No params for this service
+        # 3. (No step 3, as there's no params object to apply)
 
         # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
diff --git a/src/pipecat/services/piper/tts.py b/src/pipecat/services/piper/tts.py
index 337dcd00c..fb7b627cd 100644
--- a/src/pipecat/services/piper/tts.py
+++ b/src/pipecat/services/piper/tts.py
@@ -82,7 +82,7 @@ class PiperTTSService(TTSService):
             _warn_deprecated_param("voice_id", PiperTTSSettings, "voice")
             default_settings.voice = voice_id
 
-        # 3. No params for this service
+        # 3. (No step 3, as there's no params object to apply)
 
         # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
@@ -232,7 +232,7 @@ class PiperHttpTTSService(TTSService):
             _warn_deprecated_param("voice_id", PiperHttpTTSSettings, "voice")
             default_settings.voice = voice_id
 
-        # 3. No params for this service
+        # 3. (No step 3, as there's no params object to apply)
 
         # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
diff --git a/src/pipecat/services/resembleai/tts.py b/src/pipecat/services/resembleai/tts.py
index f0a93a182..9713cea44 100644
--- a/src/pipecat/services/resembleai/tts.py
+++ b/src/pipecat/services/resembleai/tts.py
@@ -94,7 +94,7 @@ class ResembleAITTSService(WebsocketTTSService):
             _warn_deprecated_param("voice_id", ResembleAITTSSettings, "voice")
             default_settings.voice = voice_id
 
-        # 3. No params for this service
+        # 3. (No step 3, as there's no params object to apply)
 
         # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
diff --git a/src/pipecat/services/xtts/tts.py b/src/pipecat/services/xtts/tts.py
index 4da8db72a..78ac6dfb8 100644
--- a/src/pipecat/services/xtts/tts.py
+++ b/src/pipecat/services/xtts/tts.py
@@ -124,6 +124,8 @@ class XTTSService(TTSService):
             _warn_deprecated_param("voice_id", XTTSTTSSettings, "voice")
             default_settings.voice = voice_id
 
+        # 3. (No step 3, as there's no params object to apply)
+
         # 4. Apply settings delta (canonical API, always wins)
         if settings is not None:
             default_settings.apply_update(settings)
diff --git a/tests/test_service_init.py b/tests/test_service_init.py
index 67dcbb324..377300f64 100644
--- a/tests/test_service_init.py
+++ b/tests/test_service_init.py
@@ -34,7 +34,6 @@ new services are covered automatically with no per-service maintenance.
 import importlib
 import inspect
 import pkgutil
-import warnings
 from dataclasses import fields
 
 import pytest