fix(android): unify voice speaker gating and config refresh

apps/android/app/src/main/java/ai/openclaw/android/NodeRuntime.kt

@@ -249,7 +249,12 @@ class NodeRuntime(context: Context) {
         applyMainSessionKey(mainSessionKey)
         updateStatus()
         micCapture.onGatewayConnectionChanged(true)
-        scope.launch { refreshBrandingFromGateway() }
+        scope.launch {
+          refreshBrandingFromGateway()
+          if (voiceReplySpeakerLazy.isInitialized()) {
+            voiceReplySpeaker.refreshConfig()
+          }
+        }
       },
       onDisconnected = { message ->
         operatorConnected = false
@@ -319,7 +324,7 @@ class NodeRuntime(context: Context) {
       json = json,
       supportsChatSubscribe = false,
     )
-  private val voiceReplySpeaker: TalkModeManager by lazy {
+  private val voiceReplySpeakerLazy: Lazy<TalkModeManager> = lazy {
     // Reuse the existing TalkMode speech engine (ElevenLabs + deterministic system-TTS fallback)
     // without enabling the legacy talk capture loop.
     TalkModeManager(
@@ -328,8 +333,12 @@ class NodeRuntime(context: Context) {
       session = operatorSession,
       supportsChatSubscribe = false,
       isConnected = { operatorConnected },
-    )
+    ).also { speaker ->
+      speaker.setPlaybackEnabled(prefs.speakerEnabled.value)
+    }
   }
+  private val voiceReplySpeaker: TalkModeManager
+    get() = voiceReplySpeakerLazy.value
 
   private val micCapture: MicCaptureManager by lazy {
     MicCaptureManager(
@@ -349,9 +358,7 @@ class NodeRuntime(context: Context) {
         parseChatSendRunId(response) ?: idempotencyKey
       },
       speakAssistantReply = { text ->
-        if (prefs.speakerEnabled.value) {
-          voiceReplySpeaker.speakAssistantReply(text)
-        }
+        voiceReplySpeaker.speakAssistantReply(text)
       },
     )
   }
@@ -641,6 +648,9 @@ class NodeRuntime(context: Context) {
 
   fun setSpeakerEnabled(value: Boolean) {
     prefs.setSpeakerEnabled(value)
+    if (voiceReplySpeakerLazy.isInitialized()) {
+      voiceReplySpeaker.setPlaybackEnabled(value)
+    }
   }
 
   fun refreshGatewayConnection() {

apps/android/app/src/main/java/ai/openclaw/android/voice/TalkModeManager.kt

@@ -146,6 +146,8 @@ class TalkModeManager(
   private var pendingRunId: String? = null
   private var pendingFinal: CompletableDeferred<Boolean>? = null
   private var chatSubscribedSessionKey: String? = null
+  private var configLoaded = false
+  @Volatile private var playbackEnabled = true
 
   private var player: MediaPlayer? = null
   private var streamingSource: StreamingMediaDataSource? = null
@@ -194,8 +196,21 @@ class TalkModeManager(
     }
   }
 
-  suspend fun speakAssistantReply(text: String) {
+  fun setPlaybackEnabled(enabled: Boolean) {
+    playbackEnabled = enabled
+    if (!enabled) {
+      stopSpeaking()
+    }
+  }
+
+  suspend fun refreshConfig() {
     reloadConfig()
+  }
+
+  suspend fun speakAssistantReply(text: String) {
+    if (!playbackEnabled) return
+    ensureConfigLoaded()
+    if (!playbackEnabled) return
     playAssistant(text)
   }
 
@@ -347,7 +362,7 @@ class TalkModeManager(
     lastTranscript = ""
     lastHeardAtMs = null
 
-    reloadConfig()
+    ensureConfigLoaded()
     val prompt = buildPrompt(transcript)
     if (!isConnected()) {
       _statusText.value = "Gateway not connected"
@@ -855,6 +870,12 @@ class TalkModeManager(
     return true
   }
 
+  private suspend fun ensureConfigLoaded() {
+    if (!configLoaded) {
+      reloadConfig()
+    }
+  }
+
   private suspend fun reloadConfig() {
     val envVoice = System.getenv("ELEVENLABS_VOICE_ID")?.trim()
     val sagVoice = System.getenv("SAG_VOICE_ID")?.trim()
@@ -907,6 +928,7 @@ class TalkModeManager(
       } else if (selection?.normalizedPayload == true) {
         Log.d(tag, "talk config provider=elevenlabs")
       }
+      configLoaded = true
     } catch (_: Throwable) {
       defaultVoiceId = envVoice?.takeIf { it.isNotEmpty() } ?: sagVoice?.takeIf { it.isNotEmpty() }
       defaultModelId = defaultModelIdFallback
@@ -914,6 +936,7 @@ class TalkModeManager(
       apiKey = envKey?.takeIf { it.isNotEmpty() }
       voiceAliases = emptyMap()
       defaultOutputFormat = defaultOutputFormatFallback
+      configLoaded = true
     }
   }