feat: meeting mode auto-start with sustained speech fallback

Meeting mode trigger now has two paths:
- Primary: system audio energy > 0.005 (immediate, catches most apps)
- Fallback: ~3 seconds sustained speech detection via VAD counter
  (catches Electron/WebRTC apps like Nextcloud Talk where
  ScreenCaptureKit may not capture audio)

Brief sounds or momentary speech won't trigger - only a real
conversation lasting ~3s will activate the fallback. The counter
decays during silence to prevent accumulation of brief detections.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

src-tauri/src/audio_processor.rs

@@ -46,6 +46,10 @@ pub struct AudioProcessor {
     
     // Recording Mode (voice or meeting)
     recording_mode: String,
+
+    // Meeting mode: consecutive speech detection counter.
+    // Prevents false triggers from brief sounds; requires sustained speech.
+    meeting_speech_frames: u32,
 }
 
 impl AudioProcessor {
@@ -106,6 +110,7 @@ impl AudioProcessor {
             last_event_time: std::time::Instant::now(),
             system_queue: Arc::new(Mutex::new(std::collections::VecDeque::new())),
             recording_mode,
+            meeting_speech_frames: 0,
         })
     }
 
@@ -247,29 +252,44 @@ impl AudioProcessor {
             if self.last_event_time.elapsed().as_millis() > 2000 && self.recording_mode == "meeting" {
                 if let Some(app) = &self.app_handle {
                     emit_log(app, "DEBUG", &format!(
-                        "Waiting for Meeting... SysEnergy: {:.4} (thr: 0.005), VAD Speech: {} | SysQueue empty: {}",
-                        max_system_energy, self.is_speech_active,
+                        "Waiting... SysEnergy: {:.4} (thr: 0.005), Speech: {}, SpeechFrames: {}/90 | SysQ empty: {}",
+                        max_system_energy, self.is_speech_active, self.meeting_speech_frames,
                         if let Ok(q) = self.system_queue.lock() { q.is_empty() } else { true }
                     ));
                 }
             }
 
             // MODE-SPECIFIC TRIGGER LOGIC:
-            // "voice"   -> Trigger if user speaks (VAD)
-            // "meeting" -> Trigger ONLY on system audio energy above threshold.
-            //              Speech alone NEVER triggers meeting mode to prevent
-            //              false starts when user speaks near mic without a call.
-            //              The threshold (0.005) is low enough to catch ringtones,
-            //              call audio, and notification sounds from any app.
+            // "voice"   -> Trigger if user speaks (VAD) - immediate
+            // "meeting" -> Primary: system audio energy above threshold (immediate)
+            //              Fallback: sustained speech for ~3 seconds (catches apps like
+            //              Nextcloud Talk where ScreenCaptureKit may not capture audio).
+            //              Brief speech resets the counter to prevent false triggers.
+            //
+            // The sustained speech counter uses VAD chunks (~32ms each at 16kHz/512).
+            // ~90 consecutive chunks ≈ ~3 seconds of sustained speech.
+            const MEETING_SPEECH_THRESHOLD: u32 = 90;
+
+            if self.recording_mode == "meeting" {
+                if self.is_speech_active || system_active {
+                    self.meeting_speech_frames += 1;
+                } else {
+                    // Reset if silence detected - prevents brief sounds from accumulating
+                    self.meeting_speech_frames = self.meeting_speech_frames.saturating_sub(3);
+                }
+            }
+
             let trigger = if self.recording_mode == "voice" {
                 self.is_speech_active
             } else {
-                system_active
+                // Immediate trigger on system audio energy
+                // OR sustained speech (~3s) as fallback for Electron/WebRTC apps
+                system_active || self.meeting_speech_frames >= MEETING_SPEECH_THRESHOLD
             };
 
             if trigger {
-                // Trigger Detected!
-                println!("Auto-Start: Trigger! (Mode: {}, SysEnergy: {:.4}, Speech: {})", self.recording_mode, max_system_energy, self.is_speech_active);
+                let reason = if system_active { "SystemAudio" } else { "SustainedSpeech" };
+                println!("Auto-Start: Trigger! (Mode: {}, Reason: {}, SysEnergy: {:.4}, SpeechFrames: {})", self.recording_mode, reason, max_system_energy, self.meeting_speech_frames);
                 self.waiting_for_speech = false;
 
                 // Flush Ring Buffer (Orderly: from ring_pos to end, then 0 to ring_pos)