feat: Send audioStreamEnd event for Google Live API when audio stops

- Add sendGoogleAudioStreamEnd() method to RealtimeSessionManager
- Call it when audio capture stops in RealtimeAudioInputManager
- This signals Google to flush buffered audio instead of waiting for more
- Improves responsiveness when microphone is muted or robot starts speaking

Author: studerus

app/src/main/java/ch/fhnw/pepper_realtime/manager/RealtimeAudioInputManager.kt

@@ -150,6 +150,12 @@ class RealtimeAudioInputManager(
         }
         captureThread = null
 
+        // For Google Live API: Send audioStreamEnd to flush buffered audio
+        // This signals to Google that the audio stream is paused/ended
+        if (isGoogleProvider && sessionManager.isConnected) {
+            sessionManager.sendGoogleAudioStreamEnd()
+        }
+
         cleanup()
         Log.i(TAG, "Audio capture stopped")
     }

app/src/main/java/ch/fhnw/pepper_realtime/network/RealtimeSessionManager.kt

@@ -290,6 +290,29 @@ class RealtimeSessionManager @Inject constructor() {
 
     // ==================== GOOGLE LIVE API METHODS ====================
 
+    /**
+     * Send audioStreamEnd event to Google Live API.
+     * Should be called when audio input is paused for more than ~1 second,
+     * e.g., when microphone is muted or when transitioning away from LISTENING state.
+     * This signals to Google to flush any buffered audio and process it.
+     *
+     * @return true if sent successfully
+     */
+    fun sendGoogleAudioStreamEnd(): Boolean {
+        return try {
+            val payload = JSONObject().apply {
+                put("realtimeInput", JSONObject().apply {
+                    put("audioStreamEnd", true)
+                })
+            }
+            Log.d(TAG, "Sending Google audioStreamEnd")
+            send(payload.toString())
+        } catch (e: Exception) {
+            Log.e(TAG, "Error sending Google audioStreamEnd", e)
+            false
+        }
+    }
+
     /**
      * Send an image/video frame to Google Live API via realtimeInput.video.
      * This is the streaming format that adds to context WITHOUT triggering a response.