feat: implemented forced speech and speech queue
This commit is contained in:
committed by
Pim Hutting
parent
1e77548622
commit
e51cf8fe65
@@ -1,11 +1,12 @@
|
||||
from __future__ import unicode_literals # So that we can log texts with Unicode characters
|
||||
import logging
|
||||
from threading import Thread
|
||||
|
||||
import Queue
|
||||
import zmq
|
||||
|
||||
from robot_interface.endpoints.receiver_base import ReceiverBase
|
||||
from robot_interface.state import state
|
||||
|
||||
from robot_interface.core.config import settings
|
||||
from robot_interface.endpoints.gesture_settings import GestureTags
|
||||
|
||||
@@ -32,6 +33,9 @@ class ActuationReceiver(ReceiverBase):
|
||||
self.socket.setsockopt_string(zmq.SUBSCRIBE, u"") # Causes block if given in options
|
||||
self._tts_service = None
|
||||
self._animation_service = None
|
||||
self._message_queue = Queue.Queue()
|
||||
self.message_thread = Thread(target=self._handle_messages)
|
||||
self.message_thread.start()
|
||||
|
||||
def _handle_speech(self, message):
|
||||
"""
|
||||
@@ -58,8 +62,26 @@ class ActuationReceiver(ReceiverBase):
|
||||
if not self._tts_service:
|
||||
self._tts_service = state.qi_session.service("ALTextToSpeech")
|
||||
|
||||
# Returns instantly. Messages received while speaking will be queued.
|
||||
getattr(qi, "async")(self._tts_service.say, text)
|
||||
if message.get("is_priority"):
|
||||
# Bypass queue and speak immediately
|
||||
self.clear_queue()
|
||||
self._message_queue.put(text)
|
||||
logging.debug("Force speaking immediately: {}".format(text))
|
||||
else:
|
||||
self._message_queue.put(text)
|
||||
|
||||
def clear_queue(self):
|
||||
"""
|
||||
Safely drains all pending messages from the queue.
|
||||
"""
|
||||
logging.info("Message queue size: {}".format(self._message_queue.qsize()))
|
||||
try:
|
||||
while True:
|
||||
# Remove items one by one without waiting
|
||||
self._message_queue.get_nowait()
|
||||
except Queue.Empty:
|
||||
pass
|
||||
logging.info("Message queue cleared.")
|
||||
|
||||
def _handle_gesture(self, message, is_single):
|
||||
"""
|
||||
@@ -122,6 +144,19 @@ class ActuationReceiver(ReceiverBase):
|
||||
if message["endpoint"] == "actuate/gesture/single":
|
||||
self._handle_gesture(message, True)
|
||||
|
||||
def _handle_messages(self):
|
||||
while not state.exit_event.is_set():
|
||||
try:
|
||||
text = self._message_queue.get(timeout=0.1)
|
||||
state.is_speaking = True
|
||||
self._tts_service.say(text)
|
||||
except Queue.Empty:
|
||||
state.is_speaking = False
|
||||
except RuntimeError:
|
||||
logging.error("Lost connection to Pepper. Please check if you're connected to the "
|
||||
"local WiFi and restart this application.")
|
||||
state.exit_event.set()
|
||||
|
||||
def endpoint_description(self):
|
||||
"""
|
||||
Extend the default endpoint description with gesture tags.
|
||||
|
||||
@@ -89,6 +89,7 @@ class AudioSender(SocketBase):
|
||||
try:
|
||||
while not state.exit_event.is_set():
|
||||
data = stream.read(chunk)
|
||||
if (state.is_speaking): continue # Do not send audio while the robot is speaking
|
||||
self.socket.send(data)
|
||||
except IOError as e:
|
||||
logger.error("Stopped listening: failed to get audio from microphone.", exc_info=e)
|
||||
|
||||
@@ -6,6 +6,7 @@ from robot_interface.endpoints.socket_base import SocketBase
|
||||
from robot_interface.state import state
|
||||
from robot_interface.core.config import settings
|
||||
|
||||
|
||||
class VideoSender(SocketBase):
|
||||
"""
|
||||
Video sender endpoint, responsible for sending video frames.
|
||||
|
||||
@@ -30,6 +30,7 @@ class State(object):
|
||||
self.exit_event = None
|
||||
self.sockets = []
|
||||
self.qi_session = None
|
||||
self.is_speaking = False
|
||||
|
||||
def initialize(self):
|
||||
"""
|
||||
|
||||
10
test/conftest.py
Normal file
10
test/conftest.py
Normal file
@@ -0,0 +1,10 @@
|
||||
from mock import patch, MagicMock
|
||||
|
||||
import pytest
|
||||
|
||||
|
||||
@pytest.fixture(autouse=True)
|
||||
def mock_zmq_context():
|
||||
with patch("zmq.Context") as mock:
|
||||
mock.instance.return_value = MagicMock()
|
||||
yield mock
|
||||
@@ -20,46 +20,109 @@ def zmq_context():
|
||||
yield context
|
||||
|
||||
|
||||
def test_handle_unimplemented_endpoint(zmq_context):
|
||||
def test_force_speech_clears_queue(mocker):
|
||||
"""
|
||||
Tests that the ``ActuationReceiver.handle_message`` method can
|
||||
handle an unknown or unimplemented endpoint without raising an error.
|
||||
Tests that a force speech message clears the existing queue
|
||||
and places the high-priority message at the front.
|
||||
"""
|
||||
receiver = ActuationReceiver(zmq_context)
|
||||
# Should not error
|
||||
mocker.patch("threading.Thread")
|
||||
mock_state = mocker.patch("robot_interface.endpoints.actuation_receiver.state")
|
||||
|
||||
mock_qi = mock.Mock()
|
||||
sys.modules["qi"] = mock_qi
|
||||
|
||||
mock_tts_service = mock.Mock()
|
||||
mock_state.qi_session.service.return_value = mock_tts_service
|
||||
|
||||
# Use Mock Context
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
|
||||
receiver._message_queue.put("old_message_1")
|
||||
receiver._message_queue.put("old_message_2")
|
||||
|
||||
assert receiver._message_queue.qsize() == 2
|
||||
|
||||
force_msg = {
|
||||
"endpoint": "actuate/speech",
|
||||
"data": "Emergency Notification",
|
||||
"is_priority": True,
|
||||
}
|
||||
receiver.handle_message(force_msg)
|
||||
|
||||
assert receiver._message_queue.qsize() == 1
|
||||
queued_item = receiver._message_queue.get()
|
||||
assert queued_item == "Emergency Notification"
|
||||
|
||||
|
||||
def test_handle_unimplemented_endpoint(mocker):
|
||||
"""
|
||||
Tests handling of unknown endpoints.
|
||||
"""
|
||||
mocker.patch("threading.Thread")
|
||||
|
||||
# Use Mock Context
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
|
||||
receiver.handle_message({
|
||||
"endpoint": "some_endpoint_that_definitely_does_not_exist",
|
||||
"data": None,
|
||||
})
|
||||
|
||||
|
||||
def test_speech_message_no_data(zmq_context, mocker):
|
||||
def test_speech_message_no_data(mocker):
|
||||
"""
|
||||
Tests that the message handler logs a warning when a speech actuation
|
||||
request (`actuate/speech`) is received but contains empty string data.
|
||||
Tests that if the message data is empty, the receiver returns immediately
|
||||
WITHOUT attempting to access the global robot state or session.
|
||||
"""
|
||||
mock_warn = mocker.patch("logging.warn")
|
||||
# 1. Prevent background threads from running
|
||||
mocker.patch("threading.Thread")
|
||||
|
||||
# 2. Mock the global state object
|
||||
mock_state = mocker.patch("robot_interface.endpoints.actuation_receiver.state")
|
||||
|
||||
receiver = ActuationReceiver(zmq_context)
|
||||
# 3. Create a PropertyMock to track whenever 'qi_session' is accessed
|
||||
# We attach it to the class type of the mock so it acts like a real property
|
||||
mock_session_prop = mock.PropertyMock(return_value=None)
|
||||
type(mock_state).qi_session = mock_session_prop
|
||||
|
||||
# 4. Initialize Receiver (Mocking the context to avoid ZMQ errors)
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
|
||||
# 5. Send empty data
|
||||
receiver.handle_message({"endpoint": "actuate/speech", "data": ""})
|
||||
|
||||
mock_warn.assert_called_with(mock.ANY)
|
||||
# 6. Assertion:
|
||||
# Because the code does `if not text: return` BEFORE `if not state.qi_session`,
|
||||
# the state property should NEVER be read.
|
||||
mock_session_prop.assert_not_called()
|
||||
|
||||
|
||||
def test_speech_message_invalid_data(zmq_context, mocker):
|
||||
def test_speech_message_invalid_data(mocker):
|
||||
"""
|
||||
Tests that the message handler logs a warning when a speech actuation
|
||||
request (`actuate/speech`) is received with data that is not a string (e.g., a boolean).
|
||||
Tests that if the message data is not a string, the function returns.
|
||||
:param mocker: Description
|
||||
"""
|
||||
mock_warn = mocker.patch("logging.warn")
|
||||
mocker.patch("threading.Thread")
|
||||
|
||||
receiver = ActuationReceiver(zmq_context)
|
||||
mock_state = mocker.patch("robot_interface.endpoints.actuation_receiver.state")
|
||||
|
||||
mock_session_prop = mock.PropertyMock(return_value=None)
|
||||
type(mock_state).qi_session = mock_session_prop
|
||||
|
||||
# Use Mock Context
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
|
||||
receiver.handle_message({"endpoint": "actuate/speech", "data": True})
|
||||
|
||||
mock_warn.assert_called_with(mock.ANY)
|
||||
# Because the code does `if not text: return` BEFORE `if not state.qi_session`,
|
||||
# the state property should NEVER be read.
|
||||
mock_session_prop.assert_not_called()
|
||||
|
||||
|
||||
def test_speech_no_qi(zmq_context, mocker):
|
||||
def test_speech_no_qi(mocker):
|
||||
"""
|
||||
Tests the actuation receiver's behavior when processing a speech request
|
||||
but the global state does not have an active QI session.
|
||||
@@ -69,16 +132,21 @@ def test_speech_no_qi(zmq_context, mocker):
|
||||
mock_qi_session = mock.PropertyMock(return_value=None)
|
||||
type(mock_state).qi_session = mock_qi_session
|
||||
|
||||
receiver = ActuationReceiver(zmq_context)
|
||||
mock_tts_service = mock.Mock()
|
||||
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
receiver._tts_service = mock_tts_service
|
||||
|
||||
receiver._handle_speech({"endpoint": "actuate/speech", "data": "Some message to speak."})
|
||||
|
||||
mock_qi_session.assert_called()
|
||||
receiver._tts_service.assert_not_called()
|
||||
|
||||
|
||||
def test_speech(zmq_context, mocker):
|
||||
def test_speech(mocker):
|
||||
"""
|
||||
Tests the core speech actuation functionality by mocking the QI TextToSpeech
|
||||
service and verifying that it is called correctly.
|
||||
service and verifying that the received message is put into the queue.
|
||||
"""
|
||||
mock_state = mocker.patch("robot_interface.endpoints.actuation_receiver.state")
|
||||
|
||||
@@ -89,17 +157,182 @@ def test_speech(zmq_context, mocker):
|
||||
mock_state.qi_session = mock.Mock()
|
||||
mock_state.qi_session.service.return_value = mock_tts_service
|
||||
|
||||
receiver = ActuationReceiver(zmq_context)
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
|
||||
receiver._tts_service = None
|
||||
receiver._handle_speech({"endpoint": "actuate/speech", "data": "Some message to speak."})
|
||||
|
||||
mock_state.qi_session.service.assert_called_once_with("ALTextToSpeech")
|
||||
assert receiver._message_queue.qsize() == 1
|
||||
|
||||
getattr(mock_qi, "async").assert_called_once()
|
||||
call_args = getattr(mock_qi, "async").call_args[0]
|
||||
assert call_args[0] == mock_tts_service.say
|
||||
assert call_args[1] == "Some message to speak."
|
||||
queued_item = receiver._message_queue.get()
|
||||
assert queued_item == "Some message to speak."
|
||||
|
||||
def test_speech_priority(mocker):
|
||||
"""
|
||||
Tests that a priority speech message is handled correctly by clearing the queue
|
||||
and placing the priority message at the front.
|
||||
"""
|
||||
mock_state = mocker.patch("robot_interface.endpoints.actuation_receiver.state")
|
||||
|
||||
mock_qi = mock.Mock()
|
||||
sys.modules["qi"] = mock_qi
|
||||
|
||||
mock_tts_service = mock.Mock()
|
||||
mock_state.qi_session = mock.Mock()
|
||||
mock_state.qi_session.service.return_value = mock_tts_service
|
||||
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
|
||||
receiver._message_queue.put("old_message_1")
|
||||
receiver._message_queue.put("old_message_2")
|
||||
|
||||
assert receiver._message_queue.qsize() == 2
|
||||
|
||||
priority_msg = {
|
||||
"endpoint": "actuate/speech",
|
||||
"data": "Urgent Message",
|
||||
"is_priority": True,
|
||||
}
|
||||
receiver._handle_speech(priority_msg)
|
||||
|
||||
assert receiver._message_queue.qsize() == 1
|
||||
queued_item = receiver._message_queue.get()
|
||||
assert queued_item == "Urgent Message"
|
||||
|
||||
def test_handle_messages_loop(mocker):
|
||||
"""
|
||||
Tests the background consumer loop (_handle_messages) processing an item.
|
||||
Runs SYNCHRONOUSLY to ensure coverage tools pick up the lines.
|
||||
"""
|
||||
# Patch Thread so the real background thread NEVER starts automatically
|
||||
mocker.patch("threading.Thread")
|
||||
|
||||
# Mock state
|
||||
mock_state = mocker.patch("robot_interface.endpoints.actuation_receiver.state")
|
||||
|
||||
# Setup initial speaking state to False (covers "Started speaking" print)
|
||||
mock_state.is_speaking = False
|
||||
|
||||
# Mock the TextToSpeech service
|
||||
mock_tts_service = mock.Mock()
|
||||
mock_state.qi_session.service.return_value = mock_tts_service
|
||||
|
||||
# Initialize receiver (Thread is patched, so no thread starts)
|
||||
# Use Mock Context to avoid ZMQ errors
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
|
||||
# Manually inject service (since lazy loading might handle it, but this is safer)
|
||||
receiver._tts_service = mock_tts_service
|
||||
|
||||
# This ensures the while loop iterates exactly once
|
||||
mock_state.exit_event.is_set.side_effect = [False, True]
|
||||
|
||||
# Put an item in the queue
|
||||
receiver._message_queue.put("Hello World")
|
||||
|
||||
# RUN MANUALLY in the main thread
|
||||
# This executes the code: while -> try -> get -> if print -> speaking=True -> say
|
||||
receiver._handle_messages()
|
||||
|
||||
# Assertions
|
||||
assert receiver._message_queue.empty()
|
||||
mock_tts_service.say.assert_called_with("Hello World")
|
||||
assert mock_state.is_speaking is True
|
||||
|
||||
|
||||
def test_handle_messages_queue_empty(mocker):
|
||||
"""
|
||||
Tests the Queue.Empty exception handler in the consumer loop.
|
||||
This covers the logic that resets 'state.is_speaking' to False.
|
||||
"""
|
||||
# Prevent the real background thread from starting
|
||||
mocker.patch("threading.Thread")
|
||||
|
||||
# Mock the state object
|
||||
mock_state = mocker.patch("robot_interface.endpoints.actuation_receiver.state")
|
||||
|
||||
# Setup 'is_speaking' property mock
|
||||
# We set return_value=True so the code enters the 'if state.is_speaking:' block.
|
||||
# We use PropertyMock to track when this attribute is set.
|
||||
type(mock_state).is_speaking = True
|
||||
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
|
||||
# This ensures the while loop body runs exactly once for our test
|
||||
mock_state.exit_event.is_set.side_effect = [False, True]
|
||||
|
||||
# Force get() to raise Queue.Empty immediately (simulate timeout)
|
||||
# We patch the 'get' method on the specific queue instance of our receiver
|
||||
#mocker.patch.object(receiver._message_queue, 'get', side_effect=Queue.Empty)
|
||||
|
||||
# Run the loop logic manually (synchronously)
|
||||
receiver._handle_messages()
|
||||
|
||||
# Final Assertion: Verify is_speaking was set to False
|
||||
# The code execution order is: read (returns True) -> print -> set (to False)
|
||||
# assert_called_with checks the arguments of the LAST call, which is the setter.
|
||||
assert mock_state.is_speaking is False
|
||||
|
||||
|
||||
def test_handle_messages_runtime_error(mocker):
|
||||
"""
|
||||
Tests the RuntimeError exception handler (e.g. lost WiFi connection).
|
||||
Uses a Mock ZMQ context to avoid 'Address already in use' errors.
|
||||
"""
|
||||
# Patch Thread so we don't accidentally spawn real threads
|
||||
mocker.patch("threading.Thread")
|
||||
|
||||
# Mock the state and logging
|
||||
mock_state = mocker.patch("robot_interface.endpoints.actuation_receiver.state")
|
||||
|
||||
# Use a MOCK ZMQ context.
|
||||
# This prevents the receiver from trying to bind to a real TCP port.
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
|
||||
# Initialize receiver with the mock context
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
|
||||
mock_state.exit_event.is_set.side_effect = [False, True]
|
||||
|
||||
receiver._message_queue.put("Test Message")
|
||||
|
||||
# Setup: ...BUT the service raises RuntimeError when asked to speak
|
||||
mock_tts = mock.Mock()
|
||||
mock_tts.say.side_effect = RuntimeError("Connection lost")
|
||||
receiver._tts_service = mock_tts
|
||||
|
||||
# Run the loop logic manually
|
||||
receiver._handle_messages()
|
||||
|
||||
# Assertions
|
||||
assert mock_state.exit_event.is_set.called
|
||||
|
||||
def test_clear_queue(mocker):
|
||||
"""
|
||||
Tests that the clear_queue method properly drains all items from the message queue.
|
||||
"""
|
||||
mocker.patch("threading.Thread")
|
||||
|
||||
# Use Mock Context
|
||||
mock_zmq_ctx = mock.Mock()
|
||||
receiver = ActuationReceiver(mock_zmq_ctx)
|
||||
|
||||
# Populate the queue with multiple items
|
||||
receiver._message_queue.put("msg1")
|
||||
receiver._message_queue.put("msg2")
|
||||
receiver._message_queue.put("msg3")
|
||||
|
||||
assert receiver._message_queue.qsize() == 3
|
||||
|
||||
# Clear the queue
|
||||
receiver.clear_queue()
|
||||
|
||||
# Assert the queue is empty
|
||||
assert receiver._message_queue.qsize() == 0
|
||||
|
||||
def test_gesture_no_data(zmq_context, mocker):
|
||||
receiver = ActuationReceiver(zmq_context)
|
||||
|
||||
@@ -77,7 +77,8 @@ def test_sending_audio(mocker):
|
||||
|
||||
mock_zmq_context = mock.Mock()
|
||||
send_socket = mock.Mock()
|
||||
|
||||
|
||||
mock_state.is_speaking = False
|
||||
# If there's something wrong with the microphone, it will raise an IOError when `read`ing.
|
||||
stream = mock.Mock()
|
||||
stream.read = _fake_read
|
||||
@@ -93,6 +94,36 @@ def test_sending_audio(mocker):
|
||||
send_socket.assert_called()
|
||||
|
||||
|
||||
def test_no_sending_if_speaking(mocker):
|
||||
"""
|
||||
Tests the successful sending of audio data over a ZeroMQ socket.
|
||||
"""
|
||||
mock_choose_mic = mocker.patch("robot_interface.endpoints.audio_sender.choose_mic")
|
||||
mock_choose_mic.return_value = {"name": u"Some mic", "index": 0L}
|
||||
|
||||
mock_state = mocker.patch("robot_interface.endpoints.audio_sender.state")
|
||||
mock_state.exit_event.is_set.side_effect = [False, True]
|
||||
|
||||
mock_zmq_context = mock.Mock()
|
||||
send_socket = mock.Mock()
|
||||
|
||||
mock_state.is_speaking = True
|
||||
|
||||
# If there's something wrong with the microphone, it will raise an IOError when `read`ing.
|
||||
stream = mock.Mock()
|
||||
stream.read = _fake_read
|
||||
|
||||
sender = AudioSender(mock_zmq_context)
|
||||
sender.socket.send = send_socket
|
||||
sender.audio.open = mock.Mock()
|
||||
sender.audio.open.return_value = stream
|
||||
|
||||
sender.start()
|
||||
sender.wait_until_done()
|
||||
|
||||
send_socket.assert_not_called()
|
||||
|
||||
|
||||
def _fake_read_error(num_frames):
|
||||
"""
|
||||
Helper function to simulate an I/O error during microphone stream reading.
|
||||
|
||||
Reference in New Issue
Block a user