diff --git a/tools/chatgui/src/messagedelegate.cpp b/tools/chatgui/src/messagedelegate.cpp
index e116b6ab..cc7d8307 100644
--- a/tools/chatgui/src/messagedelegate.cpp
+++ b/tools/chatgui/src/messagedelegate.cpp
@@ -23,11 +23,13 @@
 #include <QUrl>
 #include <QDebug>
 #include <QImageReader>
+#include <QPersistentModelIndex>
 
 MessageDelegate::MessageDelegate(QObject *parent)
     : QStyledItemDelegate(parent) {}
 
 struct VoicePlayState {
+    bool loading = false;
     bool playing = false;
     bool paused = false;
     qint64 startMs = 0;
@@ -48,6 +50,7 @@ static void ensureVoiceTimer() {
             bool anyActive = false;
             SoundManager* sm = SoundManager::instance();
             for (auto it = s_voiceStates.begin(); it != s_voiceStates.end(); ++it) {
+                if (it->loading) { anyActive = true; continue; }
                 if (!it->playing && !it->paused) {
                     if (it->displayFrac > 0.001f || it->displayElapsed > 0.001f) anyActive = true;
                     continue;
@@ -161,6 +164,7 @@ bool MessageDelegate::editorEvent(QEvent *event, QAbstractItemModel *model,
 
     ensureVoiceTimer();
     VoicePlayState& st = s_voiceStates[filePath];
+    if (st.loading) return true;
 
     QString durStr = index.data(MsgVoiceDurationRole).toString();
     durStr.remove('s');
@@ -175,21 +179,21 @@ bool MessageDelegate::editorEvent(QEvent *event, QAbstractItemModel *model,
 
         if (!st.playing && !st.paused) {
             /* Start playback then seek */
-            float actualDur = VoicePlayer::playOpusFile(filePath);
-            if (actualDur <= 0.0f) return false;
-            targetSec = clickFrac * actualDur;
-            st.playing = true;
-            st.paused = false;
-            st.durationSec = actualDur;
-            st.startMs = now - (qint64)(targetSec * 1000.0);
-            st.pausedPosSec = targetSec;
-            st.displayFrac = (targetSec > 0 && actualDur > 0) ? (targetSec / actualDur) : 0;
-            st.displayElapsed = targetSec;
-            int sr = SoundManager::instance()->currentPcmSampleRate();
-            if (sr > 0)
-                SoundManager::instance()->seekCurrentPcm((unsigned long long)(targetSec * sr));
+            st.loading = true;
+            QPersistentModelIndex persistent(index);
+            VoicePlayer::playOpusFile(filePath, this, [filePath, clickFrac, persistent](float actualDur) {
+                s_voiceStates[filePath].loading = false;
+                if (actualDur <= 0) return;
+                MessageDelegate::startAutoPlay(filePath, actualDur);
+                float position = clickFrac * actualDur;
+                auto& state = s_voiceStates[filePath];
+                state.startMs -= (qint64)(position * 1000); state.pausedPosSec = position;
+                state.displayFrac = clickFrac; state.displayElapsed = position;
+                int rate = SoundManager::instance()->currentPcmSampleRate();
+                if (rate > 0) SoundManager::instance()->seekCurrentPcm((unsigned long long)(position * rate));
+                if (persistent.isValid()) const_cast<QAbstractItemModel*>(persistent.model())->setData(persistent, 1, MsgVoicePlayStateRole);
+            });
             s_voiceTimer->start();
-            model->setData(index, 1, MsgVoicePlayStateRole);
         } else {
             /* Currently playing or paused — seek */
             st.startMs = now - (qint64)(targetSec * 1000.0);
@@ -246,18 +250,15 @@ bool MessageDelegate::editorEvent(QEvent *event, QAbstractItemModel *model,
         SoundManager::instance()->stopCurrentPcm();
     }
 
-    float actualDur = VoicePlayer::playOpusFile(filePath);
-    if (actualDur <= 0.0f) return false;
-    st.playing = true;
-    st.paused = false;
-    st.pausedPosSec = 0;
-    st.startMs = QDateTime::currentMSecsSinceEpoch();
-    st.durationSec = actualDur;
-    st.displayFrac = 0;
-    st.displayElapsed = 0;
-
+    st.loading = true;
+    QPersistentModelIndex persistent(index);
+    VoicePlayer::playOpusFile(filePath, this, [filePath, persistent](float actualDur) {
+        s_voiceStates[filePath].loading = false;
+        if (actualDur <= 0) return;
+        MessageDelegate::startAutoPlay(filePath, actualDur);
+        if (persistent.isValid()) const_cast<QAbstractItemModel*>(persistent.model())->setData(persistent, 1, MsgVoicePlayStateRole);
+    });
     s_voiceTimer->start();
-    model->setData(index, 1, MsgVoicePlayStateRole);
     return true;
 }
 
@@ -1492,6 +1493,7 @@ void MessageDelegate::drawAvatar(QPainter *p, const Layout &L,
 }
 
 void MessageDelegate::startAutoPlay(const QString& filePath, float actualDurationSec) {
+    if (actualDurationSec <= 0) return;
     ensureVoiceTimer();
     VoicePlayState& st = s_voiceStates[filePath];
     st.playing = true;
@@ -1506,6 +1508,8 @@ void MessageDelegate::startAutoPlay(const QString& filePath, float actualDuratio
 }
 
 void MessageDelegate::stopAllAutoPlay() {
-    for (auto it = s_voiceStates.begin(); it != s_voiceStates.end(); ++it)
-        it->playing = false;
+    VoicePlayer::cancelPending();
+    for (auto it = s_voiceStates.begin(); it != s_voiceStates.end(); ++it) {
+        it->playing = false; it->paused = false; it->loading = false;
+    }
 }
diff --git a/tools/chatgui/src/messagelist.cpp b/tools/chatgui/src/messagelist.cpp
index bd8cc11e..d746ab0a 100644
--- a/tools/chatgui/src/messagelist.cpp
+++ b/tools/chatgui/src/messagelist.cpp
@@ -960,8 +960,7 @@ void MessageList::onAttachmentDownloaded(const QString& chId, int64_t msgId) {
         if (!fp.isEmpty() && QFile::exists(fp)) {
             QTimer::singleShot(500, this, [this, fp]() {
                 MessageDelegate::stopAllAutoPlay();
-                float actualDur = VoicePlayer::playOpusFile(fp);
-                MessageDelegate::startAutoPlay(fp, actualDur);
+                VoicePlayer::playOpusFile(fp, this, [fp](float actualDur) { MessageDelegate::startAutoPlay(fp, actualDur); });
             });
         }
     }