fix: stop sending audio to Omi backend when custom STT is active (prevents listening-minute consumption)

app/lib/services/sockets/composite_transcription_socket.dart

@@ -15,6 +15,18 @@ class CompositeTranscriptionSocket implements IPureSocket {
   final String? suggestedTranscriptType;
   final String? sttProvider;
 
+  /// When true, calls to [send] route messages only to [primarySocket].
+  /// [secondarySocket] still receives forwarded transcript JSON via
+  /// [_forwardAsSuggestedTranscript] — only the raw [send] path is skipped.
+  ///
+  /// Use this when a custom STT provider handles transcription so the Omi
+  /// backend never processes raw audio — preventing listening-minute consumption.
+  ///
+  /// Note: [PureSocket] sets pingInterval=20s at connect time, so the secondary
+  /// WebSocket connection stays alive during silence without application-level
+  /// keep-alive messages.
+  final bool skipSendToSecondary;
+
   PureSocketStatus _status = PureSocketStatus.notConnected;
   IPureSocketListener? _listener;
 
@@ -26,6 +38,7 @@ class CompositeTranscriptionSocket implements IPureSocket {
     required this.secondarySocket,
     this.suggestedTranscriptType = 'suggested_transcript',
     this.sttProvider,
+    this.skipSendToSecondary = false,
   }) {
     _primaryListener = _PrimarySocketListener(this);
     _secondaryListener = _SecondarySocketListener(this);
@@ -143,7 +156,15 @@ class CompositeTranscriptionSocket implements IPureSocket {
       return;
     }
     primarySocket.send(message);
-    secondarySocket.send(message);
+    // When skipSendToSecondary is set, messages passed to send() go only to
+    // the primary socket. The secondary socket still receives forwarded
+    // transcript JSON via _forwardAsSuggestedTranscript (which calls
+    // secondarySocket.send directly), so conversation processing continues.
+    // WebSocket-level keepalives (pingInterval=20s in PureSocket) keep the
+    // secondary connection alive during silence without audio data.
+    if (!skipSendToSecondary) {
+      secondarySocket.send(message);
+    }
   }
 
   void _onPrimaryMessage(dynamic message) {

app/lib/services/sockets/transcription_service.dart

@@ -473,6 +473,11 @@ class TranscriptSocketServiceFactory {
       primarySocket: primarySocket,
       secondarySocket: secondaryService.socket,
       sttProvider: sttProvider,
+      // Custom STT handles transcription — send only forwarded transcripts to
+      // the Omi backend, never raw audio. Without this, the backend transcribes
+      // the audio stream in parallel and counts listening minutes even though
+      // the custom provider is doing the actual transcription work.
+      skipSendToSecondary: true,
     );
     return TranscriptSegmentSocketService.withSocket(
       sampleRate,