diff --git a/desktop/app.js b/desktop/app.js
index cc42f0d..483ae11 100644
--- a/desktop/app.js
+++ b/desktop/app.js
@@ -817,31 +817,56 @@ async function buildAudioPipeline(rawStream) {
       state.voice.activeNoiseFilter = 'off';
     }
   }
-  // Metering/Speaking detection
+  // Metering/Speaking detection with adaptive threshold and hysteresis.
   const analyser = ctx.createAnalyser();
-  analyser.fftSize = 512;
+  analyser.fftSize = 1024;
   processedSource.connect(analyser);
 
-  const dataArray = new Uint8Array(analyser.frequencyBinCount);
+  const timeData = new Float32Array(analyser.fftSize);
   let localIsSpeaking = false;
-  let speakCounter = 0;
+  let speechFrames = 0;
+  let silenceFrames = 0;
+  let noiseFloor = 0.004;
+  let levelEma = 0;
 
   const checkVolume = () => {
     if (!state.voice.audioContext || state.voice.audioContext.state === 'closed') return;
-    analyser.getByteFrequencyData(dataArray);
-    let sum = 0;
-    for (let i = 0; i < dataArray.length; i++) sum += dataArray[i];
-    const avg = sum / dataArray.length;
 
-    const isCurrentlySpeaking = avg > 30; // Calibrated for normal speech
+    analyser.getFloatTimeDomainData(timeData);
 
-    if (isCurrentlySpeaking) {
-      speakCounter = Math.min(speakCounter + 1, 5);
-    } else {
-      speakCounter = Math.max(speakCounter - 1, 0);
+    let sumSquares = 0;
+    let peak = 0;
+    for (let i = 0; i < timeData.length; i++) {
+      const v = timeData[i];
+      sumSquares += v * v;
+      const abs = Math.abs(v);
+      if (abs > peak) peak = abs;
     }
 
-    const newSpeakingState = speakCounter >= 2;
+    const rms = Math.sqrt(sumSquares / timeData.length);
+    const level = Math.max(rms, peak * 0.5);
+    levelEma = levelEma * 0.75 + level * 0.25;
+
+    if (!localIsSpeaking) {
+      // Learn room noise slowly while idle.
+      noiseFloor = noiseFloor * 0.98 + levelEma * 0.02;
+    } else {
+      // Do not let noise floor jump up while speaking.
+      noiseFloor = Math.min(noiseFloor, levelEma);
+    }
+
+    const startThreshold = Math.max(noiseFloor * 2.2, 0.010);
+    const stopThreshold = Math.max(noiseFloor * 1.5, 0.006);
+
+    if (levelEma > startThreshold) {
+      speechFrames = Math.min(speechFrames + 1, 8);
+      silenceFrames = 0;
+    } else if (levelEma < stopThreshold) {
+      silenceFrames = Math.min(silenceFrames + 1, 8);
+      speechFrames = Math.max(speechFrames - 1, 0);
+    }
+
+    const newSpeakingState = localIsSpeaking ? silenceFrames < 3 : speechFrames >= 2;
 
     if (newSpeakingState !== localIsSpeaking) {
       localIsSpeaking = newSpeakingState;
@@ -858,7 +883,7 @@ async function buildAudioPipeline(rawStream) {
         myVideo.parentElement.classList.toggle('speaking', localIsSpeaking);
       }
     }
-    setTimeout(checkVolume, 100);
+    setTimeout(checkVolume, 60);
   };
   checkVolume();
 
diff --git a/static/app.js b/static/app.js
index 9abc973..fa7f533 100644
--- a/static/app.js
+++ b/static/app.js
@@ -780,31 +780,56 @@ async function buildAudioPipeline(rawStream) {
     }
   }
 
-  // Metering/Speaking detection
+  // Metering/Speaking detection with adaptive threshold and hysteresis.
   const analyser = ctx.createAnalyser();
-  analyser.fftSize = 512;
+  analyser.fftSize = 1024;
   processedSource.connect(analyser);
 
-  const dataArray = new Uint8Array(analyser.frequencyBinCount);
+  const timeData = new Float32Array(analyser.fftSize);
   let localIsSpeaking = false;
-  let speakCounter = 0;
+  let speechFrames = 0;
+  let silenceFrames = 0;
+  let noiseFloor = 0.004;
+  let levelEma = 0;
 
   const checkVolume = () => {
     if (!state.voice.audioContext || state.voice.audioContext.state === 'closed') return;
-    analyser.getByteFrequencyData(dataArray);
-    let sum = 0;
-    for (let i = 0; i < dataArray.length; i++) sum += dataArray[i];
-    const avg = sum / dataArray.length;
 
-    const isCurrentlySpeaking = avg > 30; // Calibrated for normal speech
+    analyser.getFloatTimeDomainData(timeData);
 
-    if (isCurrentlySpeaking) {
-      speakCounter = Math.min(speakCounter + 1, 5);
-    } else {
-      speakCounter = Math.max(speakCounter - 1, 0);
+    let sumSquares = 0;
+    let peak = 0;
+    for (let i = 0; i < timeData.length; i++) {
+      const v = timeData[i];
+      sumSquares += v * v;
+      const abs = Math.abs(v);
+      if (abs > peak) peak = abs;
     }
 
-    const newSpeakingState = speakCounter >= 2;
+    const rms = Math.sqrt(sumSquares / timeData.length);
+    const level = Math.max(rms, peak * 0.5);
+    levelEma = levelEma * 0.75 + level * 0.25;
+
+    if (!localIsSpeaking) {
+      // Learn room noise slowly while idle.
+      noiseFloor = noiseFloor * 0.98 + levelEma * 0.02;
+    } else {
+      // Do not let noise floor jump up while speaking.
+      noiseFloor = Math.min(noiseFloor, levelEma);
+    }
+
+    const startThreshold = Math.max(noiseFloor * 2.2, 0.010);
+    const stopThreshold = Math.max(noiseFloor * 1.5, 0.006);
+
+    if (levelEma > startThreshold) {
+      speechFrames = Math.min(speechFrames + 1, 8);
+      silenceFrames = 0;
+    } else if (levelEma < stopThreshold) {
+      silenceFrames = Math.min(silenceFrames + 1, 8);
+      speechFrames = Math.max(speechFrames - 1, 0);
+    }
+
+    const newSpeakingState = localIsSpeaking ? silenceFrames < 3 : speechFrames >= 2;
 
     if (newSpeakingState !== localIsSpeaking) {
       localIsSpeaking = newSpeakingState;
@@ -821,7 +846,7 @@ async function buildAudioPipeline(rawStream) {
         myVideo.parentElement.classList.toggle('speaking', localIsSpeaking);
       }
     }
-    setTimeout(checkVolume, 100);
+    setTimeout(checkVolume, 60);
   };
   checkVolume();