diff --git a/src/robot_interface/endpoints/actuation_receiver.py b/src/robot_interface/endpoints/actuation_receiver.py
index 7fe16b7..b9435f6 100644
--- a/src/robot_interface/endpoints/actuation_receiver.py
+++ b/src/robot_interface/endpoints/actuation_receiver.py
@@ -21,6 +21,7 @@ class ActuationReceiver(ReceiverBase):
         self.create_socket(zmq_context, zmq.SUB, port)
         self.socket.setsockopt_string(zmq.SUBSCRIBE, u"")  # Causes block if given in options
         self._tts_service = None
+        self._al_memory = None
 
     def _handle_speech(self, message):
         text = message.get("data")
@@ -40,10 +41,26 @@ class ActuationReceiver(ReceiverBase):
 
         if not self._tts_service:
             self._tts_service = state.qi_session.service("ALTextToSpeech")
+        if not self._al_memory:
+            self._al_memory = state.qi_session.service("ALMemory")
+
+            # Subscribe to speech end event
+            self.status_subscriber = self._al_memory.subscriber("ALTextToSpeech/Status")  # self because garbage collect
+            self.status_subscriber.signal.connect(self._on_status_changed)
 
         # Returns instantly. Messages received while speaking will be queued.
         qi.async(self._tts_service.say, text)
 
+    @staticmethod
+    def _on_status_changed(value):  # value will contain either 'enqueued', 'started' or 'done' depending on the status
+        """Callback function for when the speaking status changes. Will change the is_speaking value of the state."""
+        if "started" in value:
+            logging.debug("Started speaking.")
+            state.is_speaking = True
+        if "done" in value:
+            logging.debug("Done speaking.")
+            state.is_speaking = False
+
     def handle_message(self, message):
         if message["endpoint"] == "actuate/speech":
             self._handle_speech(message)
diff --git a/src/robot_interface/endpoints/audio_sender.py b/src/robot_interface/endpoints/audio_sender.py
index 5cd5a6b..822bd5c 100644
--- a/src/robot_interface/endpoints/audio_sender.py
+++ b/src/robot_interface/endpoints/audio_sender.py
@@ -57,6 +57,13 @@ class AudioSender(SocketBase):
 
         try:
             while not state.exit_event.is_set():
+                # Don't send audio if Pepper is speaking
+                if state.is_speaking:
+                    if stream.is_active(): stream.stop_stream()
+                    continue
+
+                if stream.is_stopped(): stream.start_stream()
+
                 data = stream.read(chunk)
                 self.socket.send(data)
         except IOError as e:
diff --git a/src/robot_interface/main.py b/src/robot_interface/main.py
index 8874f7d..669cb82 100644
--- a/src/robot_interface/main.py
+++ b/src/robot_interface/main.py
@@ -43,8 +43,19 @@ def main_loop(context):
 
     logging.debug("Starting main loop.")
 
+    import schedule
+    test_speaking_message = {"data": "Hi, my name is Pepper, and this is quite a long message."}
+    def test_speak():
+        logging.debug("Testing speech.")
+        actuation_receiver._handle_speech(test_speaking_message)
+
+    schedule.every(10).seconds.do(test_speak)
+
     while True:
         if state.exit_event.is_set(): break
+
+        schedule.run_pending()
+
         socks = dict(poller.poll(100))
 
         for receiver in receivers:
diff --git a/src/robot_interface/state.py b/src/robot_interface/state.py
index b6f8ce1..d6e4610 100644
--- a/src/robot_interface/state.py
+++ b/src/robot_interface/state.py
@@ -18,6 +18,7 @@ class State(object):
         self.exit_event = None
         self.sockets = []  # type: List[SocketBase]
         self.qi_session = None  # type: None | ssl.SSLSession
+        self.is_speaking = False  # type: Boolean
 
     def initialize(self):
         if self.is_initialized: