FIX: auto-start voice typing after AI finishes speaking so user can respond without manual mic click. Also stop orphaned VoiceTyper instances before creating new ones to prevent mic disconnect and double-send.

src/app.js

@@ -272,6 +272,8 @@ let textMode = false;
 let voiceTyper = null;
 let config = { lb: false, allowDirect: true };
 let sessionInProgress = false;
+let _lastVoiceText = "";
+let _lastVoiceSendMs = 0;
 let autoGreetingSent = false;
 let preFetchedNews = [];
 let latestNewsUrl = null;
@@ -1017,6 +1019,8 @@ function showTextChat(show) {
 function onVoiceResult(e) {
   if (!e || !e.transcript) return;
   if (!chatInput) return;
+  // Ignore results when AI is speaking or session not in user-turn
+  if (!sessionInProgress) return;
   const text = e.transcript.trim();
   chatInput.value = text;
   if (e.isFinal && sessionInProgress) {
@@ -1025,6 +1029,10 @@ function onVoiceResult(e) {
     if (triggerRe.test(text)) {
       setCaption("Đang gửi…", "live");
       sendTextMessage();
+      // Stop voice typing after auto-send — user must click mic again to speak
+      if (voiceTyper && voiceTyper.active) {
+        voiceTyper.stop();
+      }
     } else {
       setCaption("Nhấp Send để gửi", "live");
     }
@@ -1034,6 +1042,13 @@ function onVoiceResult(e) {
 /** Start (or restart) Vietnamese voice recognition in text chat mode. */
 function startVoiceTyping() {
   if (!sessionInProgress || !chatInput) return false;
+  // Don't restart if already listening
+  if (voiceTyper && voiceTyper.active) return true;
+  // FIX: stop any previous VoiceTyper first — orphaned instances steal the mic
+  if (voiceTyper) {
+    voiceTyper.stop();
+    voiceTyper = null;
+  }
   voiceTyper = new VoiceTyper({ lang: "vi-VN", continuous: true, interimResults: true, autoSubmit: false });
   if (!voiceTyper.init()) {
     // Browser may not support Vietnamese locale — try English as fallback
@@ -1046,6 +1061,13 @@ function startVoiceTyping() {
   }
   voiceTyper.onResult = onVoiceResult;
   voiceTyper.onError = (e) => console.warn("[voiceTyper]", e);
+  voiceTyper.onSoundStart = () => setCaption("🎤 Listening…", "live");
+  voiceTyper.onSoundEnd = () => {
+    // Keep listening — only stop when AI speaks or user clicks mic again
+    if (voiceTyper && voiceTyper.active) {
+      // Do nothing; continuous listening is intentional
+    }
+  };
   voiceTyper.start();
   setCaption("🎤 Listening…", "live");
   return true;
@@ -1055,6 +1077,7 @@ function startVoiceTyping() {
 function stopVoiceTyping() {
   if (voiceTyper && voiceTyper.active) {
     voiceTyper.stop();
+    voiceTyper = null;
     setCaption("Voice stopped", "live");
   }
 }
@@ -1075,6 +1098,11 @@ function sendTextMessage() {
   const text = chatInput.value.trim();
   if (!text) return;
 
+  // Dedup: ignore if this exact message was already sent recently
+  if (text === _lastVoiceText && Date.now() - _lastVoiceSendMs < 3000) return;
+  _lastVoiceText = text;
+  _lastVoiceSendMs = Date.now();
+
   addChatMessage("user", text);
 
   // New topic? re-filter the news list and reset the image sequence.
@@ -1090,8 +1118,8 @@ function sendTextMessage() {
   if (!sendTextViaSession(text)) {
     setCaption("QUEUED…");
   }
-  // Auto-restart voice typing for continuous conversation
-  setTimeout(() => startVoiceTyping(), 300);
+  // Voice typing stays active for continuous conversation;
+  // it will auto-stop when the AI starts speaking.
 }
 
 if (newsCloseBtn) {
@@ -1363,7 +1391,18 @@ function onStatus(status) {
       break;
   }
 
+  if (status === "ai-speaking") {
+    // AI is speaking — stop voice typing so mic doesn't pick up audio
+    if (voiceTyper && voiceTyper.active) {
+      voiceTyper.stop();
+    }
+    subtitles.classList.add("visible");
+  }
+
   if (status === "user-speaking") {
+    // User's turn to speak — auto-start voice typing so user can speak
+    // immediately without clicking the mic button
+    setTimeout(() => startVoiceTyping(), 400);
     subtitles.classList.remove("visible");
     showTextChat(textMode);
   }
@@ -1660,6 +1699,10 @@ function _attachClientEvents(c) {
     }
     _topicTagsRendered = false;
     try { void renderTopicTags(); } catch (e) { console.warn("[topic] render:", e); }
+    // FIX: auto-start voice typing after AI finishes speaking so the user can respond immediately
+    if (sessionInProgress) {
+      setTimeout(() => startVoiceTyping(), 500);
+    }
   });
 
   c.addEventListener("toolcall", (e) => {