diff --git a/src/derp/mumble.py b/src/derp/mumble.py
index 47eb22b..acb3d6e 100644
--- a/src/derp/mumble.py
+++ b/src/derp/mumble.py
@@ -165,6 +165,8 @@ class MumbleBot:
         self._connect_count: int = 0
         self._sound_listeners: list = []
         self._receive_sound: bool = mu_cfg.get("receive_sound", True)
+        self._self_mute: bool = mu_cfg.get("self_mute", False)
+        self._mute_task: asyncio.Task | None = None
         self._only_plugins: set[str] | None = (
             set(mu_cfg["only_plugins"]) if "only_plugins" in mu_cfg else None
         )
@@ -225,6 +227,11 @@ class MumbleBot:
         session = getattr(self._mumble.users, "myself_session", "?")
         log.info("mumble: %s as %s on %s:%d (session=%s)",
                  kind, self._username, self._host, self._port, session)
+        if self._self_mute:
+            try:
+                self._mumble.users.myself.mute()
+            except Exception:
+                log.exception("mumble: failed to self-mute on connect")
         if self._loop:
             asyncio.run_coroutine_threadsafe(
                 self._notify_plugins_connected(), self._loop,
@@ -277,16 +284,18 @@ class MumbleBot:
     def _on_sound_received(self, user, sound_chunk) -> None:
         """Callback from pymumble thread: voice audio received.
 
-        Updates the timestamp used by the music plugin's duck monitor.
-        When this callback is registered, pymumble passes decoded PCM
-        directly and does not queue it -- no memory buildup.
+        Ignores audio from our own bots entirely -- prevents self-ducking
+        and avoids STT transcribing bot TTS/music.
         """
+        name = user["name"] if isinstance(user, dict) else None
+        bots = getattr(self.registry, "_bots", {})
+        if name and name in bots:
+            return
         prev = self._last_voice_ts
         self._last_voice_ts = time.monotonic()
         self.registry._voice_ts = self._last_voice_ts
         if prev == 0.0:
-            name = user["name"] if isinstance(user, dict) else "?"
-            log.info("mumble: first voice packet from %s", name)
+            log.info("mumble: first voice packet from %s", name or "?")
         for fn in self._sound_listeners:
             try:
                 fn(user, sound_chunk)
@@ -604,6 +613,16 @@ class MumbleBot:
         if self._mumble is None:
             return
 
+        # Unmute before streaming if self_mute is enabled
+        if self._self_mute:
+            if self._mute_task and not self._mute_task.done():
+                self._mute_task.cancel()
+                self._mute_task = None
+            try:
+                self._mumble.users.myself.unmute()
+            except Exception:
+                pass
+
         _get_vol = volume if callable(volume) else lambda: volume
         log.info("stream_audio: starting pipeline for %s (vol=%.0f%%, seek=%.1fs)",
                  url, _get_vol() * 100, seek)
@@ -822,6 +841,19 @@ class MumbleBot:
                             stderr_out.decode(errors="replace")[:500])
             if on_done is not None:
                 on_done.set()
+            # Re-mute after audio finishes
+            if self._self_mute:
+                self._mute_task = self._spawn(
+                    self._delayed_mute(3.0), name="self-mute",
+                )
+
+    async def _delayed_mute(self, delay: float) -> None:
+        """Re-mute after a delay (lets the audio buffer drain fully)."""
+        await asyncio.sleep(delay)
+        try:
+            self._mumble.users.myself.mute()
+        except Exception:
+            pass
 
     async def shorten_url(self, url: str) -> str:
         """Shorten a URL via FlaskPaste. Returns original on failure."""