diff --git a/src/AppDelegate.swift b/src/AppDelegate.swift
index b3640fa..8afd321 100644
--- a/src/AppDelegate.swift
+++ b/src/AppDelegate.swift
@@ -204,7 +204,7 @@ public final class AppDelegate: NSObject, NSApplicationDelegate {
         
         do {
             try audioRecorder.startRecording()
-            playSystemSound(name: "Hero")
+            playSystemSound(name: "Tink")
             statusBarController.updateIcon(state: .recording)
             statusBarController.buildMenu(isRecording: true)
             HUDOverlayController.shared.show(state: .recording(level: 0.1, isLagging: false, pingQuality: PingMonitor.shared.currentQuality))
@@ -217,7 +217,8 @@ public final class AppDelegate: NSObject, NSApplicationDelegate {
     }
     
     private func hasAudibleSpeech(pcmData: Data) -> Bool {
-        guard pcmData.count >= 3200 else { return false }
+        // Less than 0.25s of audio is just a momentary key tap/click
+        guard pcmData.count >= 8000 else { return false }
         var sumSquares: Double = 0
         var peak: Int16 = 0
         let count = pcmData.count / 2
@@ -232,7 +233,8 @@ public final class AppDelegate: NSObject, NSApplicationDelegate {
             }
         }
         let rms = sqrt(sumSquares / Double(count))
-        return peak > 1500 && rms > 0.010
+        // Raised thresholds: filter out mechanical keyclicks and keyboard taps
+        return peak > 2500 && rms > 0.015
     }
     
     public func stopRecording() {
@@ -243,10 +245,10 @@ public final class AppDelegate: NSObject, NSApplicationDelegate {
         let pcmData = audioRecorder.stopRecording()
         self.lastRecordedAudio = pcmData
         
-        // Fast-path: If user did NOT speak (silence / short ambient hold), dismiss immediately with ZERO delay!
+        // Fast-path: If user tapped key or did NOT speak (silence / keyclick), dismiss immediately with ZERO delay!
         let hasSpeech = hasAudibleSpeech(pcmData: pcmData)
-        if !hasSpeech && !InputBoxManager.shared.hasStreamedText {
-            print("AppDelegate: No audible speech detected. Immediate dismissal without freeze.")
+        if (!hasSpeech && !InputBoxManager.shared.hasStreamedText) || pcmData.count < 8000 {
+            print("AppDelegate: Short tap or silence detected. Immediate 0ms dismissal.")
             activeSession?.cancel()
             activeSession = nil
             isBusyFinalizing = false
@@ -308,12 +310,21 @@ public final class AppDelegate: NSObject, NSApplicationDelegate {
             print("Soniox primary finalize error:", error)
         }
         
+        // Strict Silence Gating: If server cleanly finished with empty text and no network drop,
+        // it was just silence/key tap. Dismiss immediately with zero freeze!
+        if rawText.isEmpty && !isNetworkError && !self.isNetworkDegraded {
+            print("AppDelegate: Server cleanly returned empty text for silence. Immediate dismissal.")
+            self.isBusyFinalizing = false
+            HUDOverlayController.shared.hide()
+            return
+        }
+        
         let hasSpeech = hasAudibleSpeech(pcmData: self.lastRecordedAudio)
         let isTruncated = hasSpeech && self.lastRecordedAudio.count > 64000 && rawText.count < 10
-        let needsRetry = (isNetworkError || self.isNetworkDegraded || isTruncated || rawText.isEmpty) && hasSpeech && self.lastRecordedAudio.count > 16000
+        let needsRetry = (isNetworkError || self.isNetworkDegraded || isTruncated) && hasSpeech && self.lastRecordedAudio.count > 24000
         
         if needsRetry {
-            let reason = isNetworkError ? "WebSocket network error" : (isTruncated ? "Truncated transcript (\(rawText.count) chars for \(self.lastRecordedAudio.count) bytes)" : "Empty transcript for audible speech")
+            let reason = isNetworkError ? "WebSocket network error" : "Truncated transcript (\(rawText.count) chars for \(self.lastRecordedAudio.count) bytes)"
             startSmartAudioRecovery(pcmData: self.lastRecordedAudio, reason: reason)
             return
         }